Add run_async to LlamaCppChatGenerator.

[kudos07]

Summary

Add run_async to LlamaCppChatGenerator.

• Implement run_async (wraps run() in asyncio.to_thread since llama-cpp-python is synchronous).
• Add async unit tests and an optional integration test.
• Add pytest-asyncio config and a CHANGELOG entry.

Related Issues

• fixes add run_async for LlamaCppChatGenerator #1890

Proposed Changes

• New run_async method on LlamaCppChatGenerator with identical signature to run.
• Unit tests:
  • TestLlamaCppChatGeneratorAsync::test_run_async
  • TestLlamaCppChatGeneratorAsync::test_run_async_with_params
  • TestLlamaCppChatGeneratorAsync::test_run_async_with_empty_message
• Optional integration test test_live_run_async (marked integration).
• PyTest config updated to enable asyncio mode.
• CHANGELOG updated with an Unreleased note.

How did you test it?

• Created local venv and installed dependencies (used CPU wheels for llama-cpp-python on Windows).
• Ran unit tests (mocked) locally:
  • python -m pytest tests/test_chat_generator.py::TestLlamaCppChatGeneratorAsync -v -m "not integration" — all async unit tests passed.
• Integration test included but may be skipped locally (downloads a GGUF model). CI will run full integration tests.

Notes for the reviewer

• Implementation is intentionally minimal and consistent with existing patterns (Fallback uses asyncio.to_thread).
• Streaming behavior is unchanged; run_async currently uses the same streaming semantics as run (streaming callback runs from the thread). If the project prefers a queue-based async streaming bridge (like HF local), that can be implemented in a follow-up.

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used an appropriate conventional commit type for the PR title (e.g., feat:)

[anakin87]

@kudos07 thank you for the contribution!

I'll take a look in the next few days...

[anakin87]

I left some comments on possible improvements...

[anakin87]

this file is automatically generated at release time, so please remove the addition

[anakin87]

Suggested change

	async def test_live_run_async(self, generator):
	@pytest.mark.parametrize("streaming_callback", [None, print_streaming_chunk])
	async def test_live_run_async(self, generator):

let's also verify that async+streaming works

[anakin87]

my impression is that since llama.cpp python is not thread-safe (abetlen/llama-cpp-python#951), this could be problematic.

A simple idea to fix this is the following:

1. Lock in __init__ (with an explanatory comment)
   self._inference_lock = asyncio.Lock()

2. Use the lock in run_async

    async with self._inference_lock:
        return await asyncio.to_thread(self.run, ...)

Ofc, this means only performing one generation at a time in case of multiple requests but it is thread-safe and exposes an async interface.