Add Phoenix LiveView incident room on Managed Postgres

[sdairs]

Adds a small trusted-team Phoenix LiveView incident room. Signed-in responders open incidents, append immutable timeline notes, and move investigating/monitoring/resolved state with expected versions. Ecto commits each status update and its timeline entry together; PubSub notifies browsers only after commit, while reconnect and process restart reload durable Postgres state.

Setup documents Cloud provisioning, CA/hostname verification, separate migration/runtime roles, explicit Ecto migrations and seeded BCrypt accounts. HTTP CSRF, origin checks, expiring hashed sessions and LiveView mount/event authorization remain enabled. Views show the newest 100 incidents/latest 200 entries; there is no tenant isolation or durable broadcast queue.

Validation on a dedicated ClickHouse Managed Postgres 18.6 Cloud service, 2 October 2026, native Ubuntu 24.04 ARM64:

• 3 unit and 8 real Cloud tests passed: certificate-specific wrong-CA/hostname controls, actual privilege denials, competing expected versions, blocked database sessions, Unicode/history bounds, revoked sessions and a scoped forced-entry failure proving rollback after status mutation and no broadcast.
• 8 real Chromium flow checks passed, with received WebSocket frames: two-browser delivery, own-form reset/foreign draft preservation, malformed/forged fields, concurrent status race, deliberately missed broadcast/reconnect, actual server-process restart and subsequent authorized write, native 403 without CSRF, independent replay of a revoked cookie.
• Formatting, warnings-as-errors compilation, assets build and dependency advisory audit passed. CI runs build/domain checks without Cloud credentials.

Companion article and sanitized acceptance logs/screenshot remain local for review. The dedicated Cloud fixture was deleted after independent review, with absence verified; no private credentials or endpoints are included.