perf(metrics): O(N^2) -> O(N) snapshot lookup in rolling Sharpe

get_rolling_sharpe_ratio used next((s for s in snapshots if
s.created_at == date), None) once per output row, scanning the
full snapshot list every time. With ~7600 daily rows this was
roughly 47% of recalculate_backtests() runtime in profiling.

Replaced with a one-shot {created_at: snapshot} dict lookup.
End-to-end recalculate_backtests on the 6-strategy x 11-runs
example batch: 20.8s -> 10.9s (~1.9x). recalculate_backtests
itself is unchanged and works on bundle- and legacy-loaded
backtests identically since it operates on Backtest objects.

Author: MDUYN

investing_algorithm_framework/services/metrics/sharpe_ratio.py

@@ -121,15 +121,19 @@ def get_rolling_sharpe_ratio(
     # Ensure chronological order
     snapshots = sorted(snapshots, key=lambda s: s.created_at)
 
+    # O(1) lookup of snapshot by created_at — avoids the O(N^2) scan
+    # that previously dominated recalculate_backtests on large
+    # backtests.
+    snapshot_by_dt = {s.created_at: s for s in snapshots}
+
     result = []
     for date, sharpe in rolling_sharpe_s.items():
 
         if pd.isna(sharpe):
             result.append((sharpe, date))
             continue
 
-        # Find the corresponding snapshot
-        snapshot = next((s for s in snapshots if s.created_at == date), None)
+        snapshot = snapshot_by_dt.get(date)
 
         if snapshot:
             result.append((sharpe, snapshot.created_at))