Add option to set the seed for inference

[nielstron]

For reproducibility its generally recommended to be able to set a seed for the prompt generation. This adds the option to the OpenAI-esque API endpoint and the model.

Note I am not 100% sure that setting the seed for torch is

1. enough
2. not interfering with potential parallel executions

[merrymercy]

@nielstron Could you rebase to the latest main branch?

[nielstron]

Could you rebase to the latest main branch?

Done

[merrymercy]

closed due to inactivity. Feel fee to rebase and reopen.