-----
- Add replace ``requests.Session`` with ``httpx.Client`` in all the ``tuneapi.apis`` models
- ``tuneapi.types.chats.ModelInterface`` is a class now instead of a protocol so must be initialised like
``super().__init__()``
- Make client common for a single instance of ``tuneapi.types.chats.ModelInterface`` instead of multuple clients. As is
suggested in the httpx documentation, never create multiple clients in the ``hotloop``. This has increased the speed
of ``distributed_chat`` by 8.6% in our benchmarks. It also means when running 100s of prompts in parallel in a server
this will reduce the chance of API failure.