perf: Implement Async Worker Pattern for Daemon Mode to Prevent Re-initialization

[BETAER-08]

Problem

Currently, the FileWatcher in daemon mode calls Indexer::update_file synchronously on every file change. This causes the heavy EmbeddingEngine and SQLite DB connections to be re-initialized from scratch every time. When a user saves multiple files rapidly or switches branches, this leads to severe performance bottlenecks, high memory spikes, and event processing delays.

Proposed Solution

• Separate the file watching logic from the processing logic in src/daemon/watcher.rs.
• Introduce a background Worker thread that initializes and holds a long-lived instance of EmbeddingEngine and ContextDb in memory.
• Implement an mpsc (Multi-Producer, Single-Consumer) channel to pass file change events from the Watcher to the Worker.

Definition of Done

• Watcher and Worker are separated into different threads.
• mpsc channel successfully transmits events between Watcher and Worker.
• EmbeddingEngine is loaded only once when the daemon starts and reused.
• Rapid sequential file saves do not cause the daemon to freeze or crash.