feat: activity video track (pulsing orb) for voice sessions

- ActivityVideoPublisher renders animated orb on 160x120 canvas
- Integrated into both agent.py and voice.py
- Updates confluence-collab submodule

agent.py

@@ -1,3 +1,4 @@
+import asyncio
 import os
 import json
 import base64
@@ -10,6 +11,7 @@ from livekit.plugins import openai as lk_openai, elevenlabs, silero
 import livekit.rtc as rtc
 
 from e2ee_patch import KDF_HKDF
+from activity_video import ActivityVideoPublisher
 
 logger = logging.getLogger("matrix-ai-agent")
 logging.basicConfig(level=logging.DEBUG)
@@ -103,6 +105,13 @@ async def entrypoint(ctx: JobContext):
     logger.info("Connected to room, local identity: %s", ctx.room.local_participant.identity)
     logger.info("Remote participants: %s", list(ctx.room.remote_participants.keys()))
 
+    # Publish activity video track (animated waveform bars)
+    activity_video = ActivityVideoPublisher()
+    video_track = rtc.LocalVideoTrack.create_video_track("activity", activity_video.source)
+    await ctx.room.local_participant.publish_track(video_track)
+    activity_task = asyncio.create_task(activity_video.run())
+    logger.info("Activity video track published")
+
     model = os.environ.get("LITELLM_MODEL", "claude-sonnet")
     voice_id = os.environ.get("ELEVENLABS_VOICE_ID", "21m00Tcm4TlvDq8ikWAM")
 
@@ -120,6 +129,10 @@ async def entrypoint(ctx: JobContext):
         vad=ctx.proc.userdata["vad"],
     )
 
+    @session.on("agent_state_changed")
+    def on_state_changed(ev):
+        activity_video.set_state(ev.new_state)
+
     @session.on("user_speech_committed")
     def on_speech(msg):
         logger.info("USER_SPEECH_COMMITTED: %s", msg.text_content)