[tobsch]

Fixes #352 — root cause found and fix verified against the deterministic repro (patched into the running
container's dist, reproduced with vs. without).

Root cause (confirmed by instrumenting onEvent)

Logging every onEvent call through a switch-back-at-the-boundary showed the exact sequence for the zone:

type=device_changed  isActive=false  owner=connect              -> deactivate: owner→queue, currentUri→null,
                                                                    finishTrack (stream destroyed), stopPlayback
… (inactive: position_sync / playback_state status=playing, all isActive=false → inactive branch, no-op) …
type=device_changed  isActive=true   owner=queue  status=undefined  uri=undefined   <- became active again
… then NO further ws events — only Soloist stdout `playing … [3:42 / 3:42]` heartbeats …

The reactivation arrives as a device_changed with no status. The only path that re-opens the pipe is
the status === 'playing' adopt, so device_changed slips past it, and owner === 'queue' makes onEvent
return. Mid-track this is harmless — the playing/track_changed events that follow re-adopt the zone. But
when the switch-back lands in the last seconds, Soloist hits the track end before any such event and then
emits none, so the zone is left owner=queue, stream=null, stuck at end-of-track while Soloist streams into
a pipe nobody reads.

Fix

Re-adopt on device_changed when the device is active again but has no stream:

if (event.type === 'device_changed' && !runner.stream) {
  void this.adoptConnectPlayback(zoneId, event);
  return;
}

(placed right after the !runner.ws.isActive block, so isActive is already guaranteed true there.)

Verification

Repro: two Soloist zones on one account; transfer B → seek dur-5s → transfer A.

• Without fix: zone A pins at progress == duration, is_playing == true, never advances.
• With fix (live in dist): FIX reactivation re-adopt → the spotify app took this zone over →
  playback advances (220/222s "Dass es hier so schön ist" → 0/130s "Zugabe" → continues). No freeze.

Control (switch not near the boundary) was unaffected either way.