fix(core): handle stack overflow on stm32f4 properly

The MemManage_Handler in core/embed/sys/task/stm32/systask.c is a naked assembly exception handler. It uses the EXC_RETURN bit 2 (LR & 0x4) to determine which stack was active (PSP=1, MSP=0). The original code used BEQ after TST LR,#0x4, which branches when the bit is clear, i.e., when MSP was active. The label 1f is the ‘skip stack pointer checking’ path, so the branch was taken for MSP, meaning the intended stack-overflow check for the main stack was skipped. The fix changes BEQ to BNE so the check is performed when MSP is active (the bit is clear, so Z=1, BNE not taken? Wait: TST sets Z if (LR & 0x4)==0. BNE branches if Z==0, i.e., if bit is set (PSP). That would skip for PSP. Hmm, need to re-examine. Actually the comment says ‘Skip stack ptr checking for PSP’. So label 1f is skip. We want to skip when PSP (bit set). TST LR,#0x4: if bit set, Z=0, so BNE branches -> skip. If bit clear (MSP), Z=1, BNE does not branch -> continue to check. So BNE is correct for skipping PSP. Original BEQ would skip for MSP and check for PSP. The code then compares R0 (stack pointer) against sstack. R0 was presumably loaded with the active stack pointer earlier. The comment was also updated from ‘Check if PSP is below the stack’ to ‘Check if MSP is below the stack’. This confirms the intent: detect stack overflow on the main stack (MSP) and reset MSP to the top of stack (estack). The bug meant stack overflow on MSP was not detected/rehandled, likely causing the handler to fall through to fault or infinite loop instead of recovering.