Closes #48694
Related to #38109

Problem

When the OS event queue overflows, the notify crate emits EventKind::Other as a rescan signal (e.g. macOS MustScanSubDirs, Linux IN_Q_OVERFLOW). The catch-all in fs_watcher.rs maps this to None.

Three downstream consumers silently drop events with kind: None:

• BackgroundScanner initial scan filters them out (worktree.rs)
• Settings file watcher ignores them (settings_file.rs)
• LSP didChangeWatchedFiles notifications skip them (lsp_store.rs)

The original _ => None predates these consumers. When it was written, process_events() stripped the kind and only used paths, so None was harmless. The settings watcher and LSP path were added later and assumed kind would always be set.

This is an uncommon trigger (requires event queue overflow from heavy filesystem activity) and the failure is silent, which makes it hard to link to specific user reports. But the fix is simple and the current behavior is clearly wrong.

Fix

Map the catch-all to Some(PathEventKind::Changed). "Something changed, re-read from disk" is the correct semantic for both EventKind::Other and EventKind::Any.

Extracted the mapping into a standalone path_event_kind() function with unit tests.

Release Notes:

• Fixed an issue where filesystem events could be silently dropped after OS event queue overflow, potentially causing stale settings and missing LSP notifications.