Construye el backend completo de Market Data

[GBRCenter]

@claude
Lee todos los documentos en el directorio planning, después construye el backend completo de Market Data en el directorio backend/. Esto debe incluir:

Interfaz de la API de Massive
La interfaz unificada para market data
El simulador de market data
Incluye pruebas unitarias completas.

[claude]

Claude finished @GBRCenter's task in 11m 18s —— View job • claude/issue-3-20260901-1437 • Create PR ➔

---

Done

Todo list

• Read planning/ docs and inspect current backend/app/market/ + backend/tests/market/ state
• Fix the two Massive client defects (last_trade.timestamp doesn't exist; the divisor treats nanoseconds as milliseconds, so the Massive path was writing nothing to the cache) — extracted _apply_snapshots, tests now use the real TickerSnapshot model instead of MagicMock
• Add rolling price history (600-point deque) to PriceCache + get_history()
• Add GET /api/prices/{ticker}/history endpoint (create_history_router)
• Add SSE keepalive (: ping after 15s idle) to the stream generator
• Add/extend unit tests for all of the above (cache history, Massive parsing via real model, new test_stream.py for the SSE generator + history endpoint)
• Run full test suite + lint — see limitation note below
• Update planning/MARKET_DATA_SUMMARY.md to reflect the final state
• Commit, push, and provide PR link

Summary

The market data backend (backend/app/market/) already existed on this branch — 73 passing tests per planning/MARKET_DATA_SUMMARY.md. planning/MARKET_DATA_DESIGN.md (also already in the branch) documented four concrete, verified gaps against that shipped code, which map directly onto this issue's ask. I closed all four:

1. Massive API interface — fixed two real defects: last_trade.timestamp doesn't exist on the SDK's model (the real attribute is sip_timestamp), and the divisor treated nanoseconds as milliseconds. Both were silently swallowed by a blanket except AttributeError, so the Massive path wrote nothing to the cache at all. Extracted _apply_snapshots() and rewrote test_massive.py to build snapshots via the real TickerSnapshot.from_dict(...) model instead of MagicMock — a mock answers to any attribute name, which is exactly how this shipped at 94% coverage.
2. Unified market data interface — added rolling price history to PriceCache (get_history, bounded deque, cleared on remove) so the main chart can backfill instantly.
3. Simulator — unchanged; it already matched the design doc, no gaps there.
4. Added GET /api/prices/{ticker}/history and the SSE keepalive (: ping every 15s of idle version), both needed for the frontend contract in PLAN.md §6.
5. Unit tests — added ~34 new tests (cache history, corrected Massive parsing, and a new test_stream.py covering the SSE generator and history endpoint), for roughly 94 tests total across the module.

I left the design doc's fifth gap — wiring the market module into a FastAPI lifespan — out of scope, since it depends on the DB/portfolio/watchlist routes that CLAUDE.md says are still to be built.

⚠️ Limitation: could not run tests or lint

This sandboxed environment blocks all Python execution (uv run, python3 -c, running any .py file) behind an approval step that isn't available in this headless context — only static commands like git, ls, find work. I could not run pytest or ruff to verify these changes. I manually traced through the async generator's control flow and cross-checked the Massive fix against MARKET_DATA_DESIGN.md §8 (which had already verified the real SDK's attribute names against the installed massive 2.2.0 package), but this needs a real run before merging:

cd backend
uv run --extra dev pytest -v
uv run --extra dev ruff check app/ tests/

If you'd like me to fix anything that surfaces, update --allowedTools to permit uv run / python3, or just paste the failure output back to me.