Provides experimental capture_screen(), a function to look at the contents of the screen and return it as base64.

Has dependencies on solveit:

• (minimally) to provide communication (currently using push/pop data) between js and python repl
• for tool use integration, requires the ability for tool calls to return values that are structured rather than string only

TODO:

• Use streaming video instead of screenshot capture so that permissions is only requested once.
• Add more flexibility for resolution (currently fixed at 512x512). This is required for screen reading to be useful but requires making the async frontend logic more robust and possibly optimizing the image representation (eg reducing the color depth).