fix(executor): compute p95 latency as a real 95th percentile

ExecutorMetrics.get_stats picked the p95 element with
int(n * 0.95), which collapses to the last index (the maximum)
for any batch of 20 or fewer requests, so the reported p95 was
actually the worst-case latency for typical scan batches. Use
ceil(n * 0.95) - 1, the standard nearest-rank index.

tests/unit/executor/test_concurrent.py

@@ -83,6 +83,19 @@ class TestExecutorMetrics:
         assert stats["p95_latency_ms"] >= 90.0
         assert stats["p95_latency_ms"] <= 100.0
 
+    def test_get_stats_p95_latency_small_batch(self):
+        """Test that p95 is the 95th percentile, not the maximum, for small batches."""
+        metrics = ExecutorMetrics()
+
+        # 20 requests with latencies 0ms..19ms
+        for i in range(20):
+            metrics.record_success(i * 0.001)
+
+        stats = metrics.get_stats()
+
+        # The 95th percentile of 20 samples is the 19th (18ms), not the max (19ms)
+        assert stats["p95_latency_ms"] == pytest.approx(18.0)
+
 
 class TestConcurrentExecutor:
     """Test ConcurrentExecutor functionality."""