feat(plugin-eval): implement Layer 3 Monte Carlo simulation with statistical analysis

Adds MonteCarloAnalyzer with SimResult/MonteCarloConfig dataclasses, run_simulation
async helper, and _compute_statistics using Wilson score CI, bootstrap CI,
Clopper-Pearson CI, and coefficient of variation. Wires MC layer into evaluate_skill
for Depth.DEEP and Depth.THOROUGH runs (50 and 100 runs respectively).

plugins/plugin-eval/tests/test_monte_carlo.py

@@ -0,0 +1,45 @@
+from pathlib import Path
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from plugin_eval.layers.monte_carlo import MonteCarloAnalyzer, MonteCarloConfig, SimResult
+
+
+class TestSimResult:
+    def test_sim_result(self):
+        sr = SimResult(activated=True, quality_score=0.8, tokens=2500, duration_ms=1200)
+        assert sr.activated is True
+        assert sr.errored is False
+
+
+class TestMonteCarloAnalyzer:
+    @pytest.mark.asyncio
+    @patch("plugin_eval.layers.monte_carlo.run_simulation")
+    async def test_run_with_mocked_sims(self, mock_sim, sample_skill_dir: Path):
+        mock_sim.return_value = SimResult(
+            activated=True, quality_score=0.82, tokens=2800, duration_ms=1500
+        )
+        config = MonteCarloConfig(n_runs=10, concurrency=2)
+        analyzer = MonteCarloAnalyzer(config)
+        result = await analyzer.analyze_skill(sample_skill_dir)
+        assert result.layer == "monte_carlo"
+        assert result.score > 0
+        assert "triggering" in result.sub_scores
+        assert "output_consistency" in result.sub_scores
+        assert "failure_rate" in result.sub_scores
+
+    def test_statistical_analysis(self):
+        """Test the statistical analysis on pre-computed sim results."""
+        analyzer = MonteCarloAnalyzer(MonteCarloConfig(n_runs=50))
+        results = [
+            SimResult(activated=True, quality_score=0.8 + i * 0.002, tokens=2500, duration_ms=1200)
+            for i in range(48)
+        ] + [
+            SimResult(activated=False, quality_score=0.0, tokens=500, duration_ms=200, errored=True),
+            SimResult(activated=True, quality_score=0.75, tokens=8000, duration_ms=5000),
+        ]
+        stats = analyzer._compute_statistics(results)
+        assert stats["triggering"]["activation_rate"] == pytest.approx(0.98)
+        assert stats["failure_rate"]["p_fail"] == pytest.approx(0.02)
+        assert stats["output_consistency"]["cv"] < 0.15