Make image generation more convenient to integrate

It would be more convenient for Text and richdocuments if they could obtain base64 encoded images instead of the share links that we produce.

The obvious place to make this possible is the image generation smart picker. It could post process the result of the assistant and return the base64 encoded images.

There could be other ways.

@blizzz The currently possible ways to get an image:

• smart picker: returns a list of /f/ links so you get the file IDs
• invoking the assistant: in the task output, you get the file IDs

So, from the file ID(s), you can insert the image like if it was selected in the user's storage.