- Add in ModelSettings as optional string
  - Will be added automatically to CLI args
- Set in kwargs for llama (also add to `__getstate__`) and in __init__
- Save prompt in `llama_chat_format` `chat_completion_handler`

- Set as env var rather than in CMAKE_ARGS as -DCMAKE_BUILD_TYPE - was
  seeing message in logs.
  - Can now see that being set properly.

- Needs Python 3.9 due to numpy, other deps anyway
  - Docker builder image already needs 3.9
- Add filelock as dep
  - Will use to ensure that the prompt file is not clobbered if multiple
    requests.

- Must end with `.ndjson` suffix
- Add tests

Ensure that if have multiple concurrent requests don't clobber file.

This is probably overkill since concurrent requests will actually
interrupt one another w/ default server settings.

Don't require installing all dev packages, just pyinstaller when doing `deploy.pyinstaller.mac`.

Also don't use temp var for lockfile, never referred to.