Enable live OpenRouter chat (location-aware news context)

[ShahriarNewaz]

Enable live OpenRouter chat (location-aware news context)

Summary

Chat currently runs in deterministic mock mode unless OPENROUTER_API_KEY is set and LLM_MOCK=false. We want chat to reliably use the user-selected location, fetch relevant news for that location, and produce live AI responses when configured.

Expected user behavior

1. User selects a location (e.g., “Chattogram”).
2. UI immediately shows a status like: Fetching news for Chattogram….
3. Backend performs place-scoped ingest (without mutating watchlist) and updates events for that area.
4. User sends a chat message (e.g., “Brief this place”).
5. AI reply uses the selected place’s context/events/incidents.
6. When live mode is active, the API response should indicate it is not mock (e.g., mock: false / no llm: mock).

Required configuration (keep secrets out of git)

• Set OPENROUTER_API_KEY in the local .env (do not commit).
• Set LLM_MOCK=false.
• Set INGEST_MOCK=false when verifying live ingest.

Safety

• Never present sample OSINT rows as official police reports.
• Do not leak secrets or stack traces in errors or logs.

Acceptance criteria

• With OPENROUTER_API_KEY set and LLM_MOCK=false, POST /api/chat returns a live response (not mock).
• Selecting a place triggers place-scoped ingest and chat uses the selected location.
• If ingest yields zero relevant events, chat returns a “no news found / limited context” caveat (not silent success).
• Stale selections cannot overwrite newer ones (sequence/token guard).

Manual test (after enabling live mode)

1. Put OPENROUTER_API_KEY=... in .env.
2. Set LLM_MOCK=false and INGEST_MOCK=false.
3. Start: .