Provide multiple options when triggering a prompt on rich text content

[tomusher]

When using the rich text 'magic wand' tool to apply a prompt to some text, we currently return one response and replace the content. If a user wants to try for a different response from the LLM, they need to undo and retry.

It would be a nicer experience if the user was either:

• Prompted with a customisable number of responses from the LLM which they could select from or retry for another set
• Could 'scroll' through different responses or easily retry without undoing first

[tomusher]

llm doesn't currently provide a way to request n results from an LLM. Potential to contribute upstream, or just work around that somehow.