Fix push/pull mutex lifecycle in LeaderSyncProcessor

[slashv]

Preface

Full disclosure, Codex 5.3 was used to both identify and solve this bug. I was experiencing a sync dead-lock in my LiveStore app when doing heavy bursts of events at startup. After implementing this and running my app against LiveStore with this fix it solved the problem.

I have a conceptual understanding of why this occurred as well as the solution but I have not written any code by hand nor have I examined every line in detail.

Problem

What problem does this pull request address?

• pushPullMutex could remain held on some failure/exit paths.
• Paginated pull could re-acquire before release, causing stalls/deadlock symptoms.

Solution

How does this change resolve the problem? Mention trade-offs and alternatives considered.

• Refactor local push critical section to scoped semaphore usage (withPermits(1)).
• In pull path, track lock ownership across pages and guarantee release on:
  • terminal page (NoMore)
  • chunk failure
  • interruption/stream teardown (ensuring)
• Add regression coverage for paginated non-live pull + post-pull local push progress.

Validation

What did you do to verify the change?

• direnv exec . vitest run tests/package-common/src/leader-thread/LeaderSyncProcessor.test.ts
• direnv exec . mono ts
• direnv exec . mono lint --fix

Demo (optional)

• Before: pending changes could stall indefinitely.
• After: paginated pull completes and local push resumes; regression test passes.

[schickling-assistant]

Thanks for tackling this - the mutex lifecycle changes look solid and the paginated non-live regression test is very helpful.

One coverage gap I still see: could we add tests for lock release on non-happy paths during paginated pull?

• pull chunk failure while the mutex is already held (mid-pagination)
• interruption/cancellation of the pull stream while pagination is in progress

Those seem like the most subtle regressions for this area and would give stronger confidence that + always prevent stalls.

Also, could you please rebase this branch onto latest and fix the current lint issue before merge?

Thanks again for driving this fix.

[schickling-assistant]

Small follow-up/correction (my previous comment had formatting glitches from CLI escaping):

The two symbols I referenced are releasePullMutexIfHeld and Effect.ensuring.

Request remains the same:

• please add coverage for mid-pagination pull failure and interruption/cancellation while the mutex is held
• please rebase on latest dev and fix the lint issue

Thanks again for the fix and for following up on this.