feat: exclude NEUTRAL setups from the activation gate (default on)

A NEUTRAL ("No Clear Setup") recommendation means the engine found no clear
directional trade, yet such setups could still qualify and even be crowned the
top pick purely on momentum rank (e.g. an extended momentum leader with a far,
5%-probability target). A NEUTRAL signal isn't actionable, so it shouldn't
qualify.

New `exclude_neutral` activation flag (default on): setup_qualifies drops setups
whose recommended_action is NEUTRAL. It lives in the shared gate, so it flows
through the dashboard's qualified/top-pick selection, the track record's
qualified stats, and the backtest (which computes recommended_action and gates on
meets_core). Toggleable in Admin → Settings → Activation; the frontend mirror and
activationSummary ("directional") match.

Re-run the backtest after enabling to confirm it holds/improves expectancy.

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

tests/unit/test_activation_settings.py

@@ -30,6 +30,7 @@ class TestActivationConfig:
             "min_confidence": 55.0,
             "require_high_conviction": False,
             "exclude_conflicts": False,
+            "exclude_neutral": True,
         }
 
     async def test_update_and_read_back(self, session: AsyncSession):
@@ -62,6 +63,12 @@ class TestActivationConfig:
         assert config["require_high_conviction"] is True
         assert config["exclude_conflicts"] is True
 
+    async def test_exclude_neutral_round_trip(self, session: AsyncSession):
+        # On by default; can be turned off.
+        assert (await get_activation_config(session))["exclude_neutral"] is True
+        await update_activation_config(session, {"exclude_neutral": False})
+        assert (await get_activation_config(session))["exclude_neutral"] is False
+
     async def test_rejects_negative_rr(self, session: AsyncSession):
         with pytest.raises(ValidationError):
             await update_activation_config(session, {"min_rr": -1.0})