Adding run_id to ModelRequest and ModelResponse

[adtyavrdhn]

Closes #3171

Adds a UUID run_id at agent start, thread it through GraphAgentState, and assign it to every ModelRequest/ModelResponse as they’re created so all execution paths share one identifier.

Keep run_id optional on message classes for backward compatibility.

Cover the change with tests:
(1) assert captured messages and result.all_messages() all carry the same non-empty run_id and
(2) confirm ModelMessagesTypeAdapter preserves the field through serialization.

[adtyavrdhn]

@DouweM A lot of tests will need to be modified to accommodate run_id in the snapshots.

[DouweM]

We may also need to do this in StreamedRunResult._marked_completed to get this to work right when using agent.run_stream:

pydantic-ai/pydantic_ai_slim/pydantic_ai/result.py

Lines 555 to 556 in 4cc4f35

	if message is not None:
	self._all_messages.append(message)

You'll want to test if it currently works, and if not add it.

We also definitely need to set the run_id here:

• pydantic-ai/pydantic_ai_slim/pydantic_ai/agent/abstract.py

  Lines 560 to 562 in 4cc4f35

  	# For backwards compatibility, append a new ModelRequest using the tool returns and retries
  	if parts:
  	messages.append(_messages.ModelRequest(parts))

• pydantic-ai/pydantic_ai_slim/pydantic_ai/_agent_graph.py

  Lines 742 to 744 in 4cc4f35

  	# For backwards compatibility, append a new ModelRequest using the tool returns and retries
  	if tool_responses:
  	messages.append(_messages.ModelRequest(parts=tool_responses))

[adtyavrdhn]

Yeah I made it a part of RunContext and passed it, added for the others two as well

[DouweM]

@DouweM A lot of tests will need to be modified to accommodate run_id in the snapshots.

@adtyavrdhn Yep, we'll need run_id=IsStr() everywhere.

[adtyavrdhn]

This ended up being a much bigger diff than what I was expecting

[adtyavrdhn]

Added it as a part of serialization for temporal and I added a small test for the round trip.