Add chat-level setting for the maximum output tokens of a response

[AlexVPopov]

Hey @crmne, would you accept a PR for setting the max tokens with something like with_max_output_tokens. Strictly speaking, this is an API parameter, but every major provider has this (under a different name):

• OpenAI Completions: max_completion_tokens
• OpenAI Responses: max_output_tokens
• Anthropic: max_tokens
• Gemini: max_output_tokens

Would it make sense to abstract this?