fix: start cached songs even when played by voice (autoplay + room scope)

Hope often starts an already-downloaded song from a voice command, with no
click on the player. Two things could leave it paused at 0:

- The <audio> element sits outside the LiveKit room's audio-unlock flow, so
  autoplay policy rejected the poll-driven play(). Capture the first page
  gesture (pointerdown/keydown/touchstart) and play the current src inside it
  to whitelist the element for autoplay.
- Player state is scoped per LiveKit room, but mediaScope() could return a
  stale shell room after a reconnect. Derive it from the connected room.name
  so the browser polls exactly the room the agent joined, and a "play X"
  request can never land in a scope the browser isn't watching.

Add an in-container regression test (test_media_play_resolves_cached_ready_track)
pinning the server-side resolve: an already-ready track plays without a
re-download and serves a stable audio_url.

tests/test_media.py

@@ -427,6 +427,59 @@ finally:
     assert "OK" in proc.stdout
 
 
+@needs_local_container
+def test_media_play_resolves_cached_ready_track(client):
+    """Playing an ALREADY-cached song must resolve to the on-disk ready track
+    (no redownload) and hand the browser a playable stream URL, so a cached
+    pick starts instantly rather than re-fetching from YouTube.
+
+    Regression: a cached track could be left as paused with currentTime 0 if
+    the room the agent played into diverged from the room the browser polled,
+    or if autoplay was blocked. This pins the server-side resolve + track
+    dict so a ready file is always returned with a stable audio_url."""
+    code = '''
+import sys
+sys.path.insert(0, "/opt/voice")
+import media_server as ms
+from fastapi.testclient import TestClient
+
+probe = ms.FILES_DIR / "cachetrk.webm"
+probe.write_bytes(b"CACHEDBYTES" * 40)
+ms._library["tracks"]["cachetrk"] = {
+    "id": "cachetrk", "kind": "youtube", "video_id": "cachetrk",
+    "title": "Cached Song", "channel": "Artist", "duration": 200,
+    "status": "ready", "ext": "webm", "error": None, "added_at": 1.0,
+}
+ms._start_download = lambda *a, **k: (_ for _ in ()).throw(
+    AssertionError("must not re-download an already cached track")
+)
+try:
+    c = TestClient(ms.app)
+    # play by video_id (as Hope's play_song does) must reuse the ready track
+    r = c.post("/media/api/play", json={
+        "room": "cached-test-room", "video_id": "cachetrk",
+        "title": "Cached Song", "channel": "Artist",
+    }, timeout=10)
+    assert r.status_code == 200, r.text
+    body = r.json()
+    assert body["desired"] == "playing", body
+    track = body["track"]
+    assert track["status"] == "ready", track
+    assert track["audio_url"] == "/media/api/stream/cachetrk", track
+    # the stream endpoint serves the cached bytes (browser plays it directly)
+    data = c.get("/media/api/stream/cachetrk")
+    assert data.status_code == 200
+    assert data.content == b"CACHEDBYTES" * 40
+    print("OK")
+finally:
+    probe.unlink(missing_ok=True)
+    ms._library["tracks"].pop("cachetrk", None)
+'''
+    proc = docker_exec_python(code, timeout=60, check=False)
+    assert proc.returncode == 0, proc.stderr[-800:]
+    assert "OK" in proc.stdout
+
+
 @needs_local_container
 def test_mcp_media_list_library(mcp_client_factory):
     text, is_error = mcp_client_factory(