Incorrect stack adjustment calculated for function

Version and Platform (required):

• Binary Ninja Version: 5.2.8614
• Edition: Ultimate
• OS: macOS
• OS Version: 15.5
• CPU Architecture: M1

Bug Description:
The stack adjustment on this function should be trivial to calculate but is being resolved with an incorrect value.

Stack offset of 8 on exit, subtract 4 for the return address, so it should be 4. But it is calculated as 0, confidence 255, and all callers of this function have mismatched stack pointers.

Steps To Reproduce:

1. Open this binary: Archive.zip
2. Navigate to Dog::Dog (0x00401ee0)
3. Python:

>>> current_function.stack_adjustment
OffsetWithConfidence(value=0, confidence=255)

4. Observe stack frame offset in LLIL and see that it really shouldn't be 0
5. Observe call site at _main (0x0040626a) and see that the stack misaligns right after the branches join

Expected Behavior:
I expected the stack offsets to resolve properly

Additional Information:
Seems like this is MLIL's doing? Quick step through in the debugger shows that StackAdjustmentResolver gets 4 as its estimate, but MLIL translator ignores that and goes with 0 instead.

Reproduce the issue with Archive.zip at Dog::Dog (0x00401ee0), then inspect StackAdjustmentResolver and the MLIL translator, comparing the resolver estimate with the translated stack offset. Check the LLIL frame offset and the call site in _main (0x0040626a). Done means the function reports a stack adjustment of 4 and callers no longer have mismatched stack pointers.