fix: contain watcher EMFILE runtime crashes

[Ruderud]

Summary

Closes #394.

This PR makes filesystem watcher failures recoverable instead of allowing them to cascade into a fatal Electron main-process write EPIPE dialog.

Changes:

• Pool daemon file.watch subscriptions under an existing covering watcher root to avoid duplicate full-tree chokidar watchers for the same workspace.
• Contain daemon and main-process chokidar error events, close the broken watcher, and allow a later watch request to recreate it.
• Guard local and WSL runtime stdin writes after the daemon exits so closed pipes fail as normal runtime request failures instead of uncaught EPIPE exceptions.
• Add regression tests for watcher pooling, watcher error recovery, and post-exit stdin writes.

Validation

npx vitest run src/main/ipc/fsWatchError.test.ts src/runtime/capabilities/watchPool.test.ts src/main/runtime/transports/localTransport.dispose.test.ts
npx eslint src/main/ipc/filesystem.ts src/main/ipc/fsWatchError.test.ts src/runtime/capabilities/index.ts src/runtime/capabilities/watchPool.test.ts src/main/runtime/transports/localTransport.ts src/main/runtime/transports/localTransport.dispose.test.ts src/main/runtime/transports/wslTransport.ts
npm run typecheck

[Anton-Horn]

ran the tests + typecheck locally, all green. handlers are code-agnostic so ENOSPC (the inotify limit you hit on linux for big repos) recovers too, just no test for it. fine to add later. merging.