fix(sentry): Exclude peer anomalies from shutdown summary level escalation

Peer anomalies (peer-count-high, peer-age-high) are environmental
observations common in multi-client setups, not indicators of a
problem with the shutting-down process. Exclude them from the level
escalation logic so normal shutdowns emit at info instead of warning.

Peer anomaly data is still captured in the event snapshot extra data
and tracked independently via recordMcpLifecycleAnomalyMetric. Only
process-local anomalies (high-rss, long-lived-high-rss) and cleanup
failures now escalate to warning.

Fixes XCODEBUILDMCP-Z
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: cameroncooke

src/utils/sentry.ts

@@ -402,7 +402,10 @@ export function captureMcpShutdownSummary(summary: McpShutdownSummaryEvent): voi
 
   try {
     const snapshotAnomalies = (summary.snapshot as { anomalies?: unknown }).anomalies;
-    const anomalyCount = Array.isArray(snapshotAnomalies) ? snapshotAnomalies.length : 0;
+    const PEER_ANOMALIES: ReadonlySet<string> = new Set(['peer-count-high', 'peer-age-high']);
+    const localAnomalyCount = Array.isArray(snapshotAnomalies)
+      ? snapshotAnomalies.filter((a) => typeof a === 'string' && !PEER_ANOMALIES.has(a)).length
+      : 0;
 
     const isCrashReason =
       summary.reason === 'startup-failure' ||
@@ -412,7 +415,7 @@ export function captureMcpShutdownSummary(summary: McpShutdownSummaryEvent): voi
     let level: 'error' | 'warning' | 'info';
     if (isCrashReason) {
       level = 'error';
-    } else if (summary.cleanupFailureCount > 0 || anomalyCount > 0) {
+    } else if (summary.cleanupFailureCount > 0 || localAnomalyCount > 0) {
       level = 'warning';
     } else {
       level = 'info';