Expose intelligence snapshots via API and UI

[vishr]

Problem

The intelligence detector (internal/intelligence/detector.go) runs every 60 seconds computing:

• Error rate anomalies (z-score vs baseline)
• Latency degradation detection
• Volume change detection
• Log pattern clustering
• Overall health score (0-100)

These results are only logged via slog.Info and discarded. There is no:

• API endpoint to retrieve current or historical snapshots
• Chat/AI tool exposing intelligence data
• Way for the LLM orchestrator to access anomaly insights

Current architecture

The chat UI uses an AI orchestrator with tools (status, diagnose, timeline, etc.) that query DuckDB. The intelligence detector runs independently but its analysis isn't available to any tool.

Proposal

1. Persist snapshots (in-memory ring buffer or persistent DuckDB table with Switch to persistent DuckDB for rollup tables #21)
2. Add GET /api/intelligence endpoint returning current snapshot
3. Add an AI tool (or enrich existing status tool) with intelligence data so the LLM can report anomalies proactively
4. Optionally surface health score in the chat UI sidebar

[vishr]

Housekeeping status: #27 originally added GET /api/intelligence, but #148 removed that legacy API/UI layer while retaining the detector and its in-memory LatestSnapshot. Keeping this open because the current typed observability API and MCP tool surface do not expose the snapshot. The eventual implementation should target the current REST/MCP Apps architecture rather than the paths in the original proposal.