image recognition API support

[simonff]

trafficstars

Problem

I work on the Google Earth Engine team. Our product often returns map tiles as analysis outputs, and it might be helpful to immediately send them to an an image recognition API to get a quick feel for the result.

Proposed Solution

I can always hardcode an API call to do that, but it would be nice if Jupyter had a built-in option to get a "running commentary" on image output. LLMs already can make sense of text output of the code they generate, but understanding image output would be very powerful.