refactor: simplify opportunity actions to entry/watch/avoid with confidence

- Remove dead scoring code (_score_candidate, _action_for, etc.) and
  align action decisions directly with score_opportunity_signal metrics.
- Reduce action surface from trigger/setup/chase/skip to entry/watch/avoid.
- Add confidence field (0..100) mapped from edge_score.
- Update evaluate/optimize ground-truth mapping and tests.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

tests/test_opportunity_dataset_service.py

@@ -202,7 +202,7 @@ class OpportunityEvaluationServiceTestCase(unittest.TestCase):
         self.assertEqual(result["summary"]["count"], 2)
         self.assertEqual(result["summary"]["correct"], 2)
         self.assertEqual(result["summary"]["accuracy"], 1.0)
-        self.assertEqual(result["by_action"]["trigger"]["correct"], 1)
+        self.assertEqual(result["by_action"]["entry"]["correct"], 1)
         self.assertEqual(result["trade_simulation"]["wins"], 1)
 
     def test_optimize_model_reports_recommended_weights(self):