fix: flashgrep lifecycle cleanup

[wgqqqqq]

Summary

This change tightens flashgrep lifecycle management so the workspace search daemon is owned and cleaned up through a unified core path instead of relying on desktop-only exit cleanup.

What changed

• moved flashgrep daemon cleanup responsibility into the core-owned search runtime path
• added a drop fallback for the async daemon client so leaked ownership still triggers best-effort process tree cleanup
• switched workspace search global registration from a strong global Arc to a Weak reference so the service can actually drop
• updated idle repo release scheduling to avoid keeping the search service alive unnecessarily
• replaced desktop-specific shutdown thread wiring with the shared blocking shutdown entrypoint
• unified platform process cleanup helpers around process-group or process-tree termination semantics

Why

Users reported that flashgrep could remain alive after BitFun exited. The root issue was that lifecycle ownership was not strict RAII: the search service was globally strongly referenced, and daemon cleanup depended too much on explicit desktop shutdown plumbing.

This refactor makes the ownership model tighter and gives the daemon both an explicit shutdown path and a drop-time fallback.

Impact

• desktop shutdown now uses the shared core lifecycle entrypoint
• workspace search resources can be released without being pinned forever by a global strong reference
• Windows, macOS, and Linux share the same lifecycle contract while still using platform-appropriate process cleanup primitives underneath

Validation

• cargo check --workspace
• cargo test --workspace