Move publish_task check from client to protocol egress loop

[Copilot]

The fix-holding-frames branch (commit e8e7632) attempted to prevent frame accumulation by checking publish_task in the client's frame processing loops. This broke tests that consume directly from output_queue, causing test_overlay_mode_injects_overlay_frame to hang indefinitely.

Changes

• Removed publish_task checks from client.py in _process_video_frames() and _process_audio_frames()
• Added publish_task check in protocol.py egress_loop() before enqueuing to publish_queue

Rationale

The client layer should not know about protocol-level publishing concerns. By moving the check to the protocol's egress loop:

• Frames flow through output_queue normally (tests work)
• Frames are skipped before entering publish_queue (memory issue solved)
• Proper abstraction boundaries maintained

# In protocol.py egress_loop()
async for frame in output_frames:
    # Record metrics...
    
    # Skip enqueuing when no publisher configured
    if self.publish_task is None:
        continue
        
    await asyncio.to_thread(enqueue_frame, frame)

This preserves the original intent of preventing frame accumulation while maintaining correct behavior for tests and direct output_queue consumers.

Original prompt

Reference: https://github.com/livepeer/pytrickle/actions/runs/20473748781/job/58834341289

Why is the test hanging? Is it flaky?

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.