Update 16k token limit

[stoerr]

Follow https://platform.openai.com/docs/models/gpt-3-5 : at dec 13 2023 the gpt 3.5 model will admit 16k tokens. Perhaps find that out with a request instead of using a constant...

Also , there might a conflict between the default value 1000 tokens that's used for the sidebar and the underlying system.

[stoerr]

According to https://platform.openai.com/docs/models/gpt-3-5 the gpt-3.5-turbo does not refer to the 16k window gpt-3.5-turbo-1106 yet. https://openai.com/pricing doesn't mention the older models, so we might switch to that for default.
It'd however make sense to determine the maximum token count on startup.

[stoerr]

Determining the maximum token count can be done by setting max_tokens to a gigantic value in a test request and parsing the response "This model's maximum context length is 16385 tokens."