[Feature] Can we support parameter n in OpenAI compatible API?

### Motivation

In /v1/completions and /v1/chat/completions endpoint, can we support the parameter `n`?

So that we can sampling multiple outputs for the same input.

Currently, we can only call the endpoint multiple times which is not efficient.

### Related resources

_No response_

### Additional context

_No response_

Start at the OpenAI-compatible /v1/completions and /v1/chat/completions endpoint implementations and trace how request parameters and generated outputs are handled. Done means both endpoints accept n and return multiple sampled outputs for one input without requiring repeated client calls.

Written by the indexing model from the issue text.