fix(watcher): prevent dropped events under high churn + make buffer s…

[iwayemi5]

Summary

The file watcher could silently drop events when the kernel event queue overflowed (common on busy directories or certain filesystems). This change:

• Switches to a larger, configurable in-memory event buffer
• Adds back-pressure / coalescing so we never lose create/modify/delete events
• Exposes the buffer size via CLI flag and config file
• Adds unit tests that simulate event storms

Changes

• src/watcher.rs: rewrite of the event loop using a bounded channel + coalescing map
• New --event-buffer-size flag (default 8192)
• Config support in syncer.toml
• Tests in tests/watcher_stress.rs that hammer the watcher with concurrent create/rename/delete

Test plan

• Unit tests pass (cargo test)
• Manual stress test on macOS (APFS) and Linux (ext4) with thousands of rapid file changes
• Verified no events are lost when buffer is sized appropriately
• Backward-compatible: existing configs continue to work with the new default

Closes #94

[truthixify]

In. Under-load event drops were subtle, glad you caught the root cause.