Feature: image input for VLMs

[yagil]

Equivalent to lmstudio-js’s https://lmstudio.ai/docs/typescript/llm-prediction/image-input

[ncoghlan]

Breaking down the individual pieces of this:

• Add prepare_image to the files session API
• Make the client files session attribute public
• Add a top-level prepare_image convenience API for easy interactive use
• Make the FileHandle type public
• Make the images parameter in Chat.add_user_message public, but without the implicit local file input support
• Drop the implicit local file handling from Chat instances entirely (as we're going to take a different approach to the chat session history management convenience API)

[ncoghlan]

#36 makes the APIs for adding file handles to chat history instances public.

Passing file handles to Chat.from_history/.add_entry/.append will still require the full multi-part user message format for now (it seems preferable to wait for lmstudio-ai/lmstudio-js#270 to be considered first, rather than jumping directly to independently replicating the required input types on the Python side)

[ncoghlan]

While not directly part of this issue, #37 makes the file preparation interface public, which will be used as the base for the image preparation interface.

[ncoghlan]

#38 adds the image-specific preparation APIs

[ncoghlan]

Draft SDK docs PR is up at lmstudio-ai/docs#48