fix: remove 'over-reaching' python Chat features

[cpsievert]

This PR removes the python Chat features that were deprecated in #91

Removing these features will bring the chat feature-set of the R and Python packages closer together, and significantly simplifies the Python implementation.

For context, this Chat class predates ellmer, chatlas, and shinychat. Since then, it's become more clear that these features "over-reach" into responsibilities of the AI framework (i.e., chatlas, LangChain, etc), and too tightly coupled UI with LLM state.

For reference, here is the list:

• Chat(messages=...) was removed. Use chat.ui(messages=...) instead.
• Chat(tokenizer=...) was removed. This was only relevant for .messages(token_limits=...) which was also removed.
• All parameters to .messages() were removed. This reflects an overall change philosophy for maintaining the conversation history sent to the LLM -- Chat should no longer be responsible for maintaining it -- another stateful object (perhaps the one provided by chatlas, LangChain, etc.) should be used instead. That said, .messages() is still useful if you want to access UI message state.
• The transform_user_input and transform_assistant_response decorators were removed. Instead, transformation of input/responses should be done manually and independently of Chat.
• As a result of the previous removals, .user_input(transform=...) was also removed.
• The already deprecated .set_user_message() method was officially removed (in favor of .update_user_input())

TODO

• First merge chore: deprecate 'over-reaching' Chat features #91
• Do a release with deprecations
• Maybe do a "hard deprecation" release before this?