feat: delegate lsp to handle context command filtering

[Will-ShaoHua]

Problem

In large repos (80k+ files), the @ context picker in Amazon Q's chat UI becomes unusable. The root cause is that the server sends all context command items to the webview in a single sendContextCommands notification. This triggers:

• Deep-clone overhead in mynah-ui's store (getValue() deep-clones the entire array)
• O(n) local filtering over 80k items on every keystroke
• Massive serialization cost when passing data between server and webview

Result: multi-second freezes when pressing @ or typing in the context picker.

Solution

Keep all items cached on the server. Only send a capped subset (≤1000 items) to the webview for the initial @ press. When the user types, a new filterContextCommands LSP request is sent to the server, which scores and filters all 80k items and returns the top 1000 matches.

Flow

User presses @ in mynah-ui
  → Shows capped 1000 items from store (instant)

User types "@fo"
  → Local filter runs immediately over capped 1000 items (instant feedback)
  → mynah-ui dispatches CONTEXT_COMMAND_FILTER event (debounced 200ms)
  → chat-client receives callback, sends filterContextCommands request to server
  → Server scores all 80k cached items, sorts by score, returns top ≤1000 matches
  → chat-client receives response, calls updateStore({ contextCommands: results })
  → mynah-ui re-renders the picker with server-filtered results

New file created
  → Indexer fires onContextItemsUpdated with updated list
  → processContextCommandUpdate caches full list, prioritizes new items in capped payload
  → sendContextCommands pushes updated capped set to webview
  → Server-side filter also sees the new file in the full cache

Screen.Recording.2026-03-31.at.11.25.53.AM.mov

License

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[aseemxs]

Would be great if you can attach a demo/screenshot of performan improvement or provide a build to test.

[Will-ShaoHua]

Would be great if you can attach a demo/screenshot of performan improvement or provide a build to test.

yep makes sense, provided via slack.