Local Icecast harbor for end-to-end stream testing

[anneoneone]

Problem

Going live from the dev frontend (localhost:3000 → local backend) hangs forever at "Connecting…". The panel's connect() resolves when the WebSocket opens, but the on-air cockpit only renders after the backend sends its connected frame — which it emits only once start_stream has successfully pushed to a producer target (the Icecast harbor). Locally there is no harbor, so start_stream fails or blocks and the panel sits in "Connecting…" with no error and no timeout.

Consequence: the entire Live Panel 2.0 on-air experience (spectrum analyzers, connection & upload card, RTT tile, auto-bitrate, Telegram chat card) cannot be smoke-tested locally — only the pre-air /test rehearsal and prod work today. Stale "Recording failed / N s empty" shows are a side effect of these dead local attempts.

Goal

A one-command local streaming target so a developer can exercise the full capture → MediaRecorder → WS → producer → Icecast pipeline on localhost and actually go live.

Scope

• Local Icecast (+ Liquidsoap harbor) via docker-compose mirroring the prod topology (docs/stream-rework/prod/icecast.xml, moafunk.liq) with a dev-only password. A docker compose -f … up (or a make target) should bring up /live + /test mounts reachable at a known localhost URL.
• Dev env wiring: document/scaffold the local .env producer target (ICECAST_* / STREAM_OUTPUT=icecast) so start_stream points at the local harbor instead of the retired NMS/RTMP target. (Agent can't read .env; provide an .env.example diff + README steps.)
• Preview URL: the stream spectrum + monitor need VITE_STREAM_ICECAST_TEST_URL pointing at the local mount with a valid Access-Control-Allow-Origin (Icecast-KH default). Confirm CORS works over plain http://localhost.
• README / /workflow.setup: a short "test streaming locally" section.

Sub-issue worth splitting out: go-live timeout

Independent of the harbor, the panel should not hang indefinitely. When start_stream blocks, "Connecting…" persists with no feedback. Add a go-live timeout (~15 s) that surfaces goLiveError ("still connecting — check the relay / producer target") instead of an infinite spinner. Small, self-contained frontend change in FlowOnAir.vue (handleGoLive); could ship first as a quick UX win.

Acceptance

• docker compose up (or make target) brings up a local Icecast/harbor with /live + /test
• With the documented dev .env, "Go live" on localhost reaches the on-air cockpit (not stuck at "Connecting…")
• Stream spectrum analyzer renders locally (CORS OK on the local mount)
• README documents the flow
• (optional/split) go-live timeout surfaces an error instead of an infinite spinner