Gateway auth

[benbarclay]

Running expensive services with no auth on the public internet scares me, so I got ol' mate Claude to add a basic auth system to the inference, then add some unit tests to ensure it's all good.

General idea:

• If the gateway is started with no secret, it'll run without auth (current behaviour)
• If the gateway is started with a secret, it'll verify if incoming inference requests include the secret in a bearer token

Pretty simple, but compatible with the OpenAI clients and it'll fit in trivially with the API service.