fix(vfs): keep macOS watcher setup from blocking readiness

[hongjr03]

Summary

• use PollWatcher on macOS so FSEvents run-loop startup/shutdown cannot block VFS loading
• register only deduplicated recursive roots instead of every discovered directory
• report versioned watcher readiness/failure separately from workspace scan progress
• make full LSP tests use client watched-file notifications and an explicit single worker

Root cause

The failing diagnostics and reload tests were waiting for VFS Progress(Finished). The macOS notify FSEvents backend could block while registering a watched directory before that message was emitted, leaving workspace readiness permanently false. Reducing Rust test concurrency changed the timing but did not remove that blocking path.

Validation

• cargo fmt --all -- --check
• cargo xtask check-config-artifacts
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace
• parallel cargo test -p vide --lib: 150 passed, 0 failed

The VFS regression test also records its event journal on timeout and verifies duplicate roots are registered once before loading completes.

[github-actions]

Docs preview: https://vide.pascal-lab.net/preview/pr-343/