feat: MCP server at /mcp — AI agents read server state from the panel ...

Streamable HTTP, stateless, JSON responses (proxy-friendly). Tools:
panel_overview (fleet summary), server_details (full current state),
server_history (downsampled metrics history). ASGI endpoint registered
directly on FastAPI routes (Mount misses exact /mcp) and the MCP session
manager runs in the app lifespan. mcp SDK pinned >=1.10,<2. README:
tool table, claude mcp add example, token caveat.

Co-Authored-By: Claude Code <noreply@anthropic.com>

feat: stage 1 — panel skeleton (FastAPI + SQLite) and ingest API ...

- FastAPI backend: config (pydantic-settings, GHARD_* env), aiosqlite + WAL
- SQLite schema: servers, metrics, events
- POST /api/v1/ingest: auth by X-Server-Key (sha256 hash), net rates in MB/s
  computed on panel from raw counter deltas, server card upsert
- Servers API: create (key shown once), list with dashboard summary,
  detail, PATCH name/note, delete, metrics history
- Dockerfile + docker-compose with persistent volume

Co-Authored-By: Claude Code <noreply@anthropic.com>