fix(server): accept max_completion_tokens in chat requests

[laulauraluo]

Chat requests using max_completion_tokens currently pass max_tokens=None to the backend, dropping the requested limit. Accept the new name as a validation alias while keeping the existing backend keyword, defaults and max_tokens precedence when both names are supplied.

Fixes #1907.

Validation:

• 15 model-free tests passed, including the new request-validation regressions.
• Six regular/streaming HTTP cases passed through the actual FastAPI router with a mocked model backend.
• make lint passed.

Validated on Python 3.12 with a CPU-only native build. No model inference or full model-dependent test suite was run.