feat: Room→Node Placement Map in Redis (Coordinator) #40

Goal

Persist the room → SFU node/worker mapping in Redis so every signaling pod agrees
on where a room lives.

Why this matters

With multi-pod signaling (Week 3) and the worker pool (Week 3 Day 1–2), a pod that
didn't create the room still needs to know which node/worker hosts it. The Room
Coordinator owns this placement decision (docs/ARCHITECTURE.md §2).

Tasks

• Add a Coordinator module that, on room creation, picks a node/worker and writes SET room:{id}:node
• Write room:{id}:meta (status, hostUserId, startedAt) to Redis
• On join/produce/consume, look up room:{id}:node to route correctly
• Clear the mapping when the room ends
• Handle the "room not found / node gone" case explicitly (client gets a clear error → re-create)
• Add a TTL or reconciliation so stale mappings don't linger after a crash

Acceptance criteria

• A pod that didn't create the room can still resolve its node from Redis
• Mapping is removed on room end
• Stale mappings are reconciled (no orphan room→node entries)

Notes

This is the multi-pod prerequisite for graceful SFU drain in Week 7 Day 4.