Add drift-comparator pushdown for conditional market fetches

[IamJasonBian]

Summary

Implements a drift-comparator layer that conditionally fetches market data only when needed, reducing API calls from 8+ per cycle to 1 in flat markets.

• DriftComparator: New module caching per-symbol price/metrics state; triggers full OHLCV fetches on price drift (>0.5%, tighter in high-vol) or staleness (>30min)
• Pushdown in run_once(): Multi-quote first (1 API call), full-fetch only drifted symbols, use cached metrics for others
• refresh_open_order_metrics(): On-demand method to refresh metrics for symbols with open orders
• Tests: 14 new unit tests validating threshold logic, caching, and edge cases

Verification

• All 14 new tests pass
• No new test failures (7 pre-existing failures unrelated to these changes)
• Cold start behavior unchanged: all symbols get full fetch on first cycle
• Subsequent cycles use cached metrics if price hasn't drifted

[IamJasonBian]

Code Review — PR #51

Overall: Strong optimization — reducing API calls from 8+ to 1 in flat markets is a meaningful improvement. The DriftComparator is well-designed with clean separation and thorough tests (14 unit + integration tests).

Issues

1. No fallback if get_multi_quote fails — If the multi-quote call raises or returns empty, quotes will be empty/None and the entire symbol loop will skip full fetches (or crash on quotes.get(symbol)). Add a try/except that falls back to full-fetch-all:

   try:
       quotes = self.data_provider.get_multi_quote(self.symbols)
   except Exception:
       quotes = {}
       needs_fetch = list(self.symbols)  # fallback: fetch everything

2. Cached path skips state_manager market data — When using cached metrics, self.state_manager.update_metrics(symbol, metrics) is called, but the state manager may also depend on market_data (e.g., for daily bars used in drift metrics). Verify that _compute_drift_metrics() still works correctly when some symbols only have cached metrics.

3. drift_metrics computed before symbol loop but also used by it — _compute_drift_metrics() reads cached daily bars, but those bars only exist after fetch_market_data runs. On the first cycle this is fine (all symbols do full fetch), but on subsequent cycles the drift metrics will be stale for symbols using cached data. Consider if that's acceptable.

4. refresh_open_order_metrics duplicates fetch logic — The full-fetch-or-cache pattern in refresh_open_order_metrics is identical to the main loop. Consider extracting a _fetch_or_cache_metrics(symbol, price) helper to avoid the duplication.

Strengths

• Staleness window (30 min) is a sensible safety net
• High-vol regime using tighter thresholds is smart
• Rate limiting only on full-fetch symbols saves time
• Integration test with Netlify blob format is thorough

---

Generated by Claude Code

[IamJasonBian]

Comparator might not be the best use case here - reviewing later. Let's implement for another use case