fix: improve reconnection stability and fix resource leaks

[Yana-Hangabina]

Summary

Fixes several bugs in the socket reconnection flow that cause crashes, lost file sync operations, and resource leaks. These issues affect all users but are more visible on unstable connections (closes #309).

Problems

1. _resolveById throws when VFS root is uninitialized — during reconnect, this.root is briefly undefined. Any call to _resolveById (from ClientManager.removePosition, cursor tracking, etc.) throws EntryNotFound, crashing the 500ms status timer chain. All callers already handle undefined returns via optional chaining.

2. SCMCollectionProvider destroyed on every reconnect — initializingPromise unconditionally disposes and recreates both ClientManager and SCMCollectionProvider after each successful joinProject. This kills in-progress LocalReplicaSCMProvider.overwrite() (file sync), discards file watchers, and resets collaboration state.

3. connectionRejected listener leak in joinProject — uses .on() instead of .once(), accumulating a new listener on every joinProject call. After ~10 reconnects, triggers MaxListenersExceededWarning.

4. EventBus.on('socketioConnectedEvent') leak — updateEventHandlers registers a new EventBus listener for onConnectionAccepted on every call without removing the previous one.

5. 5-second socket emit timeout — too aggressive for connections with non-trivial latency (proxy, remote servers). joinDoc calls frequently time out, cascading into disconnects.

6. error handler throws uncaught exceptions — initInternalHandlers throws on socket errors, which can crash the extension host.

Changes

File	Change
src/core/remoteFileSystemProvider.ts	_resolveById returns undefined when root is not set; skip SCM/ClientManager recreation when they already exist
src/collaboration/clientManager.ts	Add reconnect() method to re-register handlers on new socket; save handlers as named object; guard removePosition in updateStatus forEach
src/api/socketio.ts	Increase emit/joinProject timeout to 15s; use .once() for connectionRejected; dispose previous EventBus listener before re-registering; log errors instead of throwing

Test plan

• Open a project, verify file sync completes without "Local Replica creation failed"
• Disconnect and reconnect network, verify status bar recovers and collaboration state is preserved
• Check developer console for absence of MaxListenersExceededWarning and EntryNotFound errors
• Verify real-time collaboration (cursor tracking, user list) works after reconnect