perf(server): reuse position/rotation arrays in network entity sync

[RZDESIGN]

Summary

Physics runs at 60 Hz but network synchronization runs at 30 Hz (every 2 ticks, controlled by TICKS_PER_NETWORK_SYNC). Each tick, Entity.checkAndEmitUpdates() fires UPDATE_POSITION / UPDATE_ROTATION events for every moving entity, which the NetworkSynchronizer handles by allocating a new [x, y, z] or [x, y, z, w] array and assigning it to the entity sync object.

Since the sync object persists across the 2-tick window, the first tick's array is silently discarded when the second tick overwrites it — creating a throwaway array that immediately becomes garbage.

Before

entitySync.p = [ payload.position.x, payload.position.y, payload.position.z ];
entitySync.r = [ payload.rotation.x, payload.rotation.y, payload.rotation.z, payload.rotation.w ];

Every call allocates a new array, even when one already exists on the sync object.

After

const p = entitySync.p;
if (p) {
  p[0] = payload.position.x;
  p[1] = payload.position.y;
  p[2] = payload.position.z;
} else {
  entitySync.p = [ payload.position.x, payload.position.y, payload.position.z ];
}

Reuse the existing array when present (in-place index writes); only allocate on the first update in each sync cycle.

Impact

With 100 moving entities, this eliminates ~6,000 short-lived arrays per second (100 entities × 2 arrays × 30 wasted allocations/s). On busy worlds with 200+ entities the savings double. This directly reduces minor GC frequency and pause time on the server.

Changes

File	Change
server/src/networking/NetworkSynchronizer.ts	_onEntityUpdatePosition: reuse existing .p array via index writes
server/src/networking/NetworkSynchronizer.ts	_onEntityUpdateRotation: reuse existing .r array via index writes

Risk Assessment

• Correctness: The arrays still contain the exact same values — the serializer (msgpackr) reads indices [0], [1], [2] (and [3] for quaternions), which are identical whether the array is new or reused.
• Sync lifecycle: The sync object (and its arrays) is created fresh each sync cycle via _createEntitySync → { i: entity.id! }. On first position update in a cycle, the else branch allocates. On subsequent ticks in the same cycle, the if branch reuses. After _clearSyncQueue, everything is discarded cleanly.
• No observable difference: The only change is when the array object is allocated, not what it contains.

Test Plan

• Run a world with 100+ moving entities — verify all entity positions/rotations sync correctly to connected clients
• Monitor server GC (Bun --smol or V8 --trace-gc) — expect fewer minor GC cycles
• Stress test with rapid position changes (physics-heavy scene) — verify no stale data leaks across sync cycles