[speed] pre-fetch AI response as you type

AI responses are slow. So prefetch responses, similar to link prefetching*, as you are typing. That way, when you type a message, proof read it, and hit send, the response shows up immediately cause it was fetched seconds ago.

Use case: This could be great for local AI's or cheap models.

Considerations

• don't automatically trigger plugins for pre-fetched responses (as it could perform write-operations), but be ready to trigger them when hitting enter.

• close previous connection when fetching new prefetch request.

Probably want some level of customizations:

• enable/disable

• debouncing time in milliseconds

• https://www.wikiwand.com/en/Link_prefetching