Fix concurrent map write in db.sync() method

[corylanou]

Summary

• Fixed concurrent map write panic in db.sync() method by adding missing mutex locks
• Fixed PageSize race condition in SnapshotReader() method
• Added race test to prevent regression
• Issue: Concurrent map write panic in TestStore_Integration #696

Problem 1: Concurrent Map Write (Critical)

The db.maxLTXFileInfos.m map was being accessed without mutex protection in two locations:

• Line 893: delete(db.maxLTXFileInfos.m, 0)
• Lines 898-904: Writing to db.maxLTXFileInfos.m[0]

This caused sporadic "fatal error: concurrent map writes" panics during tests.

Problem 2: PageSize Race (Minor)

The db.pageSize field was being read directly in SnapshotReader() without synchronization while it could be written during init(). This is unlikely to occur in practice since:

• pageSize is only written once during initialization
• It's an int which is typically atomic on most architectures
• The value never changes after being set

However, it's technically a race according to Go's memory model, and the fix is trivial (use the existing thread-safe PageSize() method), so it's worth fixing for correctness.

Solution

1. Added mutex Lock/Unlock calls around the unprotected map operations
2. Changed SnapshotReader() to use db.PageSize() instead of direct field access

Test plan

• Added TestDB_ConcurrentMapWrite to reproduce the race conditions
• Verified test detects races when run with go test -race
• Confirmed fixes resolve both race conditions
• All existing tests pass

🤖 Generated with Claude Code