[MEDIUM] Fix memory leaks and cleanup issues in React hooks

[NateIsern]

Description

Several React hooks and services have memory leak risks due to missing cleanup, unbounded data structures, and race conditions with async operations.

Affected Files

• packages/frontend/hooks/usePresence.ts:22, 43, 46-55
• packages/frontend/services/socketService.ts:565-567, 949-1050

Problems

1. usePresence: setState on unmounted component (usePresence.ts:22)

socketService.getPresence(userId)
  .then(setIsOnline)     // Can fire after component unmounts
  .catch(() => setIsOnline(false));

No AbortController or cleanup flag to prevent setState after unmount.

2. usePresenceBulk: unstable dependency causes re-subscriptions (usePresence.ts:46-55)

useEffect(() => {
  // Creates new subscriptions...
  return () => {
    unsubscribes.forEach(unsubscribe => unsubscribe());
  };
}, [userIds.join(',')]);  // String join creates new value each render

userIds.join(',') creates a new string reference on each render, potentially causing infinite re-subscription loops.

3. Socket service listener Maps grow unbounded (socketService.ts:949-1050)

presenceListeners and followListeners Maps have a MAX of 500 entries, but if consumers forget to unsubscribe, the map keeps growing until the limit. Unsubscribe functions may never be called for components that unmount without cleanup.

4. Feed update queue unbounded growth (socketService.ts:565-567)

if (currentQueue.length > this.MAX_BATCH_SIZE * 2) {
  this.feedUpdateQueue.set(feedType, posts.slice(-this.MAX_BATCH_SIZE));
}

Queue can grow to MAX_BATCH_SIZE * 2 before any trimming occurs.

Expected Behavior

1. Add cleanup flags or AbortController to all async operations in useEffect
2. Use useMemo or useRef for stable dependency references
3. Enforce strict listener limits with automatic cleanup of oldest entries
4. Trim feed queue more aggressively or use a fixed-size ring buffer

Impact

• Severity: Medium
• Risk: Gradual memory growth, React warnings about setState on unmounted components, potential performance degradation over time