Add WHISPER_COMPUTE_TYPE and other CPU optomizations

[toddejohnson]

Add WHISPER_COMPUTE_TYPE env var so it can be forced. Also auto detected CPU and default to int8 to work around the below error:

/app/whisper_server.py:35: DeprecationWarning:
        on_event is deprecated, use lifespan event handlers instead.

        Read more about it in the
        [FastAPI docs for Lifespan Events](https://fastapi.tiangolo.com/advanced/events/).

  @app.on_event("startup")
INFO:     Started server process [7]
INFO:     Waiting for application startup.
22:41:08  INFO  Loading Whisper model 'large-v3' on device 'cpu' (pid 7)...
22:41:09  INFO  HTTP Request: GET https://huggingface.co/api/models/Systran/faster-whisper-large-v3/revision/main "HTTP/1.1 200 OK"
ERROR:    Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/starlette/routing.py", line 638, in lifespan
    async with self.lifespan_context(app) as maybe_state:
  File "/usr/local/lib/python3.10/dist-packages/fastapi/routing.py", line 241, in __aenter__
    await self._router._startup()
  File "/usr/local/lib/python3.10/dist-packages/fastapi/routing.py", line 4884, in _startup
    await handler()
  File "/app/whisper_server.py", line 48, in load_whisper_model
    model = WhisperModel(
  File "/usr/local/lib/python3.10/dist-packages/faster_whisper/transcribe.py", line 689, in __init__
    self.model = ctranslate2.models.Whisper(
ValueError: Requested float16 compute type, but the target device or backend do not support efficient float16 computation.

ERROR:    Application startup failed. Exiting.
Starting Whisper server: model=large-v3 device=cpu workers=1