test: Failover Test

[Harxhit]

Goal

Verify durable state and clean recovery behavior, then run the weekly review.

Tasks

• Kill a backend pod mid-stream; confirm clients get a clear error and can re-create/re-join (per §13: media is non-durable, RTO seconds)
• Verify streams/viewer_sessions docs reflect reality after the crash (statuses closed out or reconciled)
• Confirm room→node mapping is cleaned up for the dead node
• Run a leak soak: 100+ connect/disconnect cycles, confirm flat resource usage
• Confirm Mongo readiness gating behaves when Mongo is bounced
• Fill in review.md and move unfinished work to its Backlog

Acceptance criteria

• Clean client recovery after a node loss
• No resource leak across a soak test
• Persistence reflects post-crash reality
• Week 4 review completed; backlog captured

Then

Open the Week 4 review.