fix: populate early-warning/combined on the latest snapshot + recent history

The early-warning score showed n/a because it required an exact date match
between the live benchmark (Alpaca, may have today's bar) and the stored
universe breadth (DB, often a day behind), which blanked the newest snapshot —
the one the UI displays.

- Look up the divergence as-of the snapshot date (newest value within a 7-day
  lag) instead of requiring an exact match.
- Backfill early_warning + combined onto recent existing snapshots (the index
  history predates this signal) so the 7/30-day trends populate on the first run
  rather than only filling in over the coming weeks.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

tests/unit/test_regime_monitor.py

@@ -53,6 +53,15 @@ def test_attach_early_warning_none_falls_back_to_index():
     assert result["combined"]["score"] == 80.0  # no early warning -> just the index
 
 
+def test_divergence_asof_tolerates_small_lag():
+    from app.services.regime_monitor_service import _divergence_asof
+    items = [(date(2026, 6, 1), 55.0), (date(2026, 6, 3), 60.0)]
+    assert _divergence_asof(items, date(2026, 6, 3)) == 60.0   # exact date
+    assert _divergence_asof(items, date(2026, 6, 4)) == 60.0   # 1-day lag -> newest
+    assert _divergence_asof(items, date(2026, 6, 20)) is None  # too stale
+    assert _divergence_asof([], date(2026, 6, 3)) is None
+
+
 # ---------------------------------------------------------------------------
 # Price sub-scores
 # ---------------------------------------------------------------------------