Electrum

tests: try to make "backup" regtest less flaky

Kind of shooting in the dark, but looking at logs, when the "backup" test fails,
Alice logs "Sending CHANNEL_REESTABLISH" and "Sending ERROR", but
Bob never receives them, instead he logs "Disconnecting: LightningPeerConnectionClosed()".
Maybe Alice's TCP socket should just be flushed, as done here.

submarine_swaps: wait for broadcast in wait_for_htlcs_and_broadcast

Wait for broadcast of the funding transaction in
`SwapManager.wait_for_htlcs_and_broadcast` before returning.
Previously it might have returned the swap funding txid before the
funding tx was broadcast.
This caused the regtest `test_swapserver_success_forward` to fail
as it expected the funding tx to be broadcast once the CLI command
returned.

See:
```
.***** test_swapserver_success_forward ******
initializing alice
funding alice
ec354953f96e0de7be4354ab156611da9e31bd85c2d25664bdb995950530c12e
initializing bob
funding bob
1df19a419b5ae5b961bae6cfaf4e7c8e772c6632d30782fab8de6992a8901e4a
mining 1 blocks
starting daemon (PID 7471)
/tmp/alice/regtest/wallets/default_wallet
true
starting daemon (PID 7502)
/tmp/bob/regtest/wallets/default_wallet
true

alice opens channel
mining 3 blocks
wait until alice sees channel open.
wait until alice sees channel open..
wait until alice sees channel open...
alice initiates forward-swap
{
"lightning_amount": "0.01967661",
"onchain_amount": "0.02",
"txid": "b0f9ba48b670ee4a1f182655fe18f3f7d5c310923b7c85ad396b24e0a2f8e0cc"
}
utxo b0f9ba48b670ee4a1f182655fe18f3f7d5c310923b7c85ad396b24e0a2f8e0cc:0 does not exist
FDaemon stopped
Daemon stopped

======================================================================
FAIL: test_swapserver_success_forward (tests.regtest.TestLightningSwapserver.test_swapserver_success_forward)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/tmp/cirrus-ci-build/tests/regtest.py", line 105, in test_swapserver_success_forward
self.run_shell(['swapserver_success_forward'])
File "/tmp/cirrus-ci-build/tests/regtest.py", line 19, in run_shell
assert process.returncode == 0
^^^^^^^^^^^^^^^^^^^^^^^
AssertionError
```

Can be reproduced with:
```diff
diff --git a/electrum/submarine_swaps.py b/electrum/submarine_swaps.py
index f83d653f9..242aef070 100644
--- a/electrum/submarine_swaps.py
+++ b/electrum/submarine_swaps.py
@@ -1052,6 +1052,7 @@ class SwapManager(Logger):
@log_exceptions
async def broadcast_funding_tx(self, swap: SwapData, tx: Transaction) -> None:
swap.funding_txid = tx.txid()
+ await asyncio.sleep(5)
await self.network.broadcast_transaction(tx)

async def reverse_swap(
```

tests: lnhelpers: (trivial) mark method private, add type hints

tests: move create_test_channels to lnhelpers.py

test_lnwallet: unittest channel rebalancing

Add some unittest coverage for the channel rebalance flow.

Co-authored-by: SomberNight <somber.night@protonmail.com>

LNWallet: don't allow rebalancing through frozen channels

Don't allow rebalancing through a channel if it is frozen for the
direction required for the rebalance.
If this is allowed by the gui it creates opaque failures as then
pathfinding potentially fails downstream.

lnworker: fix rebalance with frozen channels

Don't exclude r_tags for frozen channels from the route creation in
create_route_for_single_htlc if the start_node of the routing hint
is unequal to our node id.

When doing a rebalance us (chan_1) -> bob -> us (chan_2, frozen for sending),
we would exclude the invoice r_tag for chan_2 because the chan
is our chan and frozen for sending,
resulting in us being unable to find a route back to us trough bob.

This is a regression from #9692 (964ffbd2).

tests: ln graph_definition: allow multiple chans between peers

tests: split off helpers from test_lnpeer into new lnhelpers.py

lnwatcher: introduce loop to trigger callbacks

Introduce a taskgroup and polling loop to LNWatcher to guarantee
the callbacks get called at least once every
LNWatcher.MAX_CALLBACK_TRIGGER_DELAY_SEC (10 min).
This should prevent callbacks that operate on time instead of
blockheight from becoming (very) stale if there are no blockchain
events triggering the callbacks for a longer time.
Not entirely set about the 10 min delay, might as well be 5/2/1 min?

util: dedup now() helper function

Deduplicate the now() helper function declared in 3 different modules
into electrum.util.
Use it consistently in submarine_swaps.py.
Cleanup imports of lnchannel.py.

LNGossip: put maintain_db on thread

I regularly see asyncio (debug) warnings that maintain_db
is blocking the event loop on my relatively fast laptop.

E.g. 3 warnings during a single session:
```
2.41 | I | channel_db.ChannelDB | Deleting 903 old policies
2.68 | I | channel_db.ChannelDB | Deleting 3954 orphaned channels
2.68 | W | asyncio | Executing <Task pending name='Task-22' coro=<LNGossip.maintain_db() running at /var/home/user/code/vibecoding_vm/electrum/electrum/lnworker.py:728> wait_for=<Future pending cb=[Task.task_wakeup()] created at /usr/lib64/python3.14/asyncio/base_events.py:459> cb=[set.discard(), TaskGroup._on_done()] created at /var/home/user/code/vibecoding_vm/electrum/electrum/util.py:1756> took 0.335 seconds
...
242.71 | I | channel_db.ChannelDB | Deleting 151 old policies
243.69 | I | channel_db.ChannelDB | Deleting 11819 orphaned channels
243.69 | W | asyncio | Executing <Task pending name='Task-22' coro=<LNGossip.maintain_db() running at /var/home/user/code/vibecoding_vm/electrum/electrum/lnworker.py:728> wait_for=<Future pending cb=[Task.task_wakeup()] created at /usr/lib64/python3.14/asyncio/base_events.py:459> cb=[set.discard(), TaskGroup._on_done()] created at /var/home/user/code/vibecoding_vm/electrum/electrum/util.py:1756> took 1.010 seconds
...
363.72 | I | channel_db.ChannelDB | Deleting 108 old policies
363.90 | I | channel_db.ChannelDB | Deleting 2922 orphaned channels
363.90 | W | asyncio | Executing <Task pending name='Task-22' coro=<LNGossip.maintain_db() running at /var/home/user/code/vibecoding_vm/electrum/electrum/lnworker.py:728> wait_for=<Future pending cb=[Task.task_wakeup()] created at /usr/lib64/python3.14/asyncio/base_events.py:459> cb=[set.discard(), TaskGroup._on_done()] created at /var/home/user/code/vibecoding_vm/electrum/electrum/util.py:1756> took 0.205 seconds
```

tests: lnchannel: also test with SRK chan type

fix regression from https://github.com/spesmilo/electrum/pull/10622,
where the default for `TEST_ANCHOR_CHANNELS` was flipped from False to True.

test_lnchannel: unittest should_be_closed_due_to_expiring_htlcs

lnchannel: give offered htlcs some time to fail on restart

Wait for 30 seconds after restart before force closing channels
due to timed-out htlcs we offered to the peer. Maybe the peer
will fail them offchain once we reestablished the connection.
This should prevent some unnecessary force closes.
We could skip the whole loop but this seems more readable.

Note: it might take some time until the lnwatcher callback calls
should_be_closed_due_to_expiring_htlcs again, however the same issue
affects the existing 30sec timeout above as well.

add builder keys for svanstaa

new 'stored_at' syntax

backported from levelDB branch

ci: security review: log effort level

If the model is logged it makes sense to log the effort level as well.

ci: security review: pass prompt as system prompt

By appending the prompt to the system prompt it is prioritized over
the actual user prompt (the diff), so it apparently is more persistent
in the context. This e.g. could improve the output if a large diff
is provided as the actual task (system prompt) is treated with higher
priority than the large diff itself.

ci: security review: give more permissions to claude

With `--dangerously-skip-permissions` claude can also execute
commands. I thought the `-p` flag for headless usage already
did that, but apparently not
(looking at https://code.claude.com/docs/en/headless).

As it runs in the isolated CI container this seems fine and
might make it more powerful (e.g. allowing it to run and
modify unittests to verify behavior).

tests: regtest: make test_just_in_time less flaky

The regtest just_in_time is flaky on the CI because it tries to open
two channels right after each other (mining 3 blocks in between).
If the channel opener (LSP/Bob) hasn't caught up yet and is still
on the same height as the previous channel open it will refuse to open
another channel due to the anchor channel key derivation limitation.

This adds a helper to pause the test until bob has caught up to the newly
mined 3 blocks before attempting the second open.

LNWallet: make get_invoice_features base feature independent

Makes the LNWallet._get_invoice_features method modify any
given base features instead of constructing the base
features internally. This way it can be used independently
of bolt11/12.

lnutil: update LN_FEATURES_IMPLEMENTED

Add missing LNWALLET_FEATUERS features to LN_FEATURES_IMPLEMENTED.

lnutil: add to_tlv_bytes() to LnFeatures

lnutil: make dependencies context dependent

Make transitive dependencies of lightning features context dependent.
This allows to validate different transitive dependencies for the same
feature in different contexts.
For example BASIC_MPP_* depends on payment_secret for bolt 11 invoices,
but has no dependencies when being used in the context of bolt 12 invoices.