Problem

Manual force-push paths updated BinSync's in-memory master state, but they did not reliably flush those changes through the client commit/update path. That made force pushes appear to succeed in the UI while other environments did not receive the changes.

This also exposed two related sync issues:

• schedule_job() returned early whenever auto-commit was disabled, which also blocked manual work that still needs to be scheduled explicitly.
• Function sync could abort before applying address-keyed comments when setting the function itself was a no-op or returned a failure through the remote client.
• Function force-push did not include comments inside the pushed function range, even though comments are stored as address-keyed artifacts separate from Function.

Solution

• Add an explicit _flush_force_push_state() helper and call it after function/global/type/segment force-push updates.
• Keep manual scheduled jobs runnable even when auto-commit is disabled.
• Isolate function set no-op/failure handling so address-keyed comments can still sync.
• Force-push comments that fall inside pushed function ranges and normalize raw string comments returned by decompiler APIs into Comment artifacts.
• Add controller-level tests for persistence, raw string comments, no-op function set behavior, and manual scheduling.

Verification

• conda run -n ghidra python -m pytest tests -q --ignore=tests/test_angr_gui.py -> 20 passed, 1 warning
• conda run -n ghidra python -m compileall -q binsync tests
• git diff --check

Notes

This branch is clean against upstream main and combines the related force-push/comment/scheduler fixes so reviewers do not have to reason about a stacked force-push series.