fix(voice): propagate ChatMessage.interrupted through proto serializer

[toubatbrian]

Summary

_chat_item_to_proto in livekit-agents/livekit/agents/voice/remote_session.py omits the interrupted field when converting llm.ChatMessage to its proto form. The model has the field (interrupted: bool = False in livekit-agents/livekit/agents/llm/chat_context.py:320) and the proto schema has it (ChatMessage.interrupted at field 4 in livekit_agent_session.proto), but the serializer drops it on the floor. Result: every conversation_item_added event on the wire reports interrupted=false, even for messages whose model copy was committed with interrupted=True.

Repro

Drive the agent into a partial-speech-then-interrupt scenario — e.g. ask for a long story, then send a new runInput after ~2 s of speaking. Subscribe to AgentSessionMessage.event over the lk.agent.session byte-stream topic. The framework's AgentActivity commits the partial assistant message with interrupted=True (model-side), but the emitted proto event has item.message.interrupted=false on the wire.

After this PR:

// conversation_item_added event for the interrupted partial speech
{
  "item": {
    "message": {
      "role": "ASSISTANT",
      "content": [{ "text": "Alright, picture this: three" }],
      "interrupted": true, // <-- now correctly propagated
    },
  },
}

Fix

         pb_msg = agent_pb.ChatMessage(
             id=item.id,
             role=pb_role,
             content=content,
+            interrupted=item.interrupted,
             metrics=_metrics_to_proto(item.metrics),
         )

Single field added to the ChatMessage constructor. No API change, no behaviour change for the model — only the wire serialization is affected.

Test plan

• Verified locally against examples/voice_agents/basic_agent.py with an external proto consumer. Before the fix, all messages had interrupted=false. After the fix, the partial assistant message that was interrupted mid-speech shows interrupted=true, matching the JS framework's behaviour.
• CI: make check (ruff lint + mypy) passes.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.