miner: fix `addPackageTxs` unsigned integer overflow

What changed, and why it matters

This commit fixes a math bug in the Bitcoin Core block-building code. The original code subtracted a safety margin from the maximum block weight and then compared the current block weight to that lower number. Because the variables are unsigned integers, if the block weight was already larger than expected, the subtraction could underflow and wrap around to a huge number, making the comparison behave incorrectly. The fix rewrites the comparison as an addition on the other side, which cannot underflow. The practical effect is on when the miner decides to stop trying to add more transactions; the bug could make it give up too early or keep going too long in edge cases.

Security signals we found

Evidence from the diff

Changed components

diff --git a/src/node/miner.cpp b/src/node/miner.cpp
index a08c70e2..e75bc3a6 100644
--- a/src/node/miner.cpp
+++ b/src/node/miner.cpp
@@ -397,8 +397,8 @@ void BlockAssembler::addPackageTxs(int& nPackagesSelected, int& nDescendantsUpda
 
             ++nConsecutiveFailed;
 
-            if (nConsecutiveFailed > MAX_CONSECUTIVE_FAILURES && nBlockWeight >
-                    m_options.nBlockMaxWeight - BLOCK_FULL_ENOUGH_WEIGHT_DELTA) {
+            if (nConsecutiveFailed > MAX_CONSECUTIVE_FAILURES && nBlockWeight +
+                    BLOCK_FULL_ENOUGH_WEIGHT_DELTA > m_options.nBlockMaxWeight) {
                 // Give up if we're close to full and haven't succeeded in a while
                 break;
             }