fix: address metrics review — real p50, seconds, _sum/_count

- Add p50_ms to EndpointStats via shared _percentile() helper
- Use real p50 instead of avg for quantile="0.5" (avg ≠ median)
- Rename metric to _seconds with values in seconds (Prometheus convention)
- Add _sum and _count lines to summary for rate() query support
- Rename cache_total → cache_lookups_total for clarity

src/veilrender/routes/metrics.py

@@ -55,29 +55,36 @@ def _build_metrics() -> str:
 
     # Counters — cache
     lines = [
-        "# HELP veilrender_cache_total Cache hits and misses by endpoint.\n",
-        "# TYPE veilrender_cache_total counter\n",
+        "# HELP veilrender_cache_lookups_total Cache lookups by endpoint and result.\n",
+        "# TYPE veilrender_cache_lookups_total counter\n",
     ]
     for name, ep in _ENDPOINTS:
         lines.append(
-            f'veilrender_cache_total{{endpoint="{name}",result="hit"}} {ep.cache_hits}\n'
+            f'veilrender_cache_lookups_total{{endpoint="{name}",result="hit"}} {ep.cache_hits}\n'
         )
         lines.append(
-            f'veilrender_cache_total{{endpoint="{name}",result="miss"}} {ep.cache_misses}\n'
+            f'veilrender_cache_lookups_total{{endpoint="{name}",result="miss"}} {ep.cache_misses}\n'
         )
     parts.append("".join(lines))
 
-    # Latency
+    # Latency — summary with quantiles, _sum, _count
     lines = [
-        "# HELP veilrender_request_duration_ms Request latency in milliseconds.\n",
-        "# TYPE veilrender_request_duration_ms summary\n",
+        "# HELP veilrender_request_duration_seconds Request latency in seconds.\n",
+        "# TYPE veilrender_request_duration_seconds summary\n",
     ]
     for name, ep in _ENDPOINTS:
+        count = ep.successes + ep.failures
         lines.append(
-            f'veilrender_request_duration_ms{{endpoint="{name}",quantile="0.5"}} {ep.avg_ms:.1f}\n'
+            f'veilrender_request_duration_seconds{{endpoint="{name}",quantile="0.5"}} {ep.p50_ms / 1000:.6f}\n'
         )
         lines.append(
-            f'veilrender_request_duration_ms{{endpoint="{name}",quantile="0.95"}} {ep.p95_ms:.1f}\n'
+            f'veilrender_request_duration_seconds{{endpoint="{name}",quantile="0.95"}} {ep.p95_ms / 1000:.6f}\n'
+        )
+        lines.append(
+            f'veilrender_request_duration_seconds_sum{{endpoint="{name}"}} {ep.total_ms / 1000:.6f}\n'
+        )
+        lines.append(
+            f'veilrender_request_duration_seconds_count{{endpoint="{name}"}} {count}\n'
         )
     parts.append("".join(lines))
 

src/veilrender/stats.py

@@ -43,15 +43,24 @@ class EndpointStats:
         total = self.successes + self.failures
         return self.total_ms / total if total > 0 else 0.0
 
-    @property
-    def p95_ms(self) -> float:
-        """95th percentile latency in milliseconds."""
+    def _percentile(self, q: float) -> float:
+        """Compute a percentile from the latency window."""
         if not self._latencies:
             return 0.0
         sorted_lat = sorted(self._latencies)
-        idx = int(len(sorted_lat) * 0.95)
+        idx = int(len(sorted_lat) * q)
         return sorted_lat[min(idx, len(sorted_lat) - 1)]
 
+    @property
+    def p50_ms(self) -> float:
+        """50th percentile (median) latency in milliseconds."""
+        return self._percentile(0.50)
+
+    @property
+    def p95_ms(self) -> float:
+        """95th percentile latency in milliseconds."""
+        return self._percentile(0.95)
+
     @property
     def success_rate(self) -> float:
         """Success rate as a percentage."""