Lazy-load chat histories and serve lightweight list summaries (avoid full global index scan on every page load)

[bradjin8]

Summary

UI testing is blocked by slow chat loading: the home page rescans Cursor's full global KV index across all projects on every load (often 1–2+ min), and opening a workspace assembles every conversation with full bubble bodies before the sidebar renders. There is no summary cache.

Problem

• GET /api/workspaces runs list_workspace_projects(), which loads all composerData, all bubbleId rows, and all messageRequestContext rows from global storage on every request—even though the UI only needs project cards (id, name, conversationCount, lastModified).
• GET /api/workspaces/<id>/tabs runs assemble_workspace_tabs(), which loads all global bubbles then builds full tab payloads for every conversation in the project before first paint.
• No application-level cache for derived summaries.

Proposed solution

1. Summary-only list endpoints (no full global bubbleId scan on list paths).
2. Lazy-load full conversation content when a tab is selected (split API + workspace UI).
3. Optional mtime-keyed cache for summaries.

Acceptance criteria

• Home and workspace sidebar usable in < 3 s on a large local fixture.
• List routes do not full-scan global bubbles (test or documented audit).
• Export/full assembly paths remain correct.
• Tests + optional warnings for partial/stale summary builds.