Writable documents

[gsproston-scottlogic]

As part of the indirect injection R&D effort, we should allow the user to write to the document store to put prompts in a document that the LLM then interprets.

Things to consider:

• This will change the document system from read-only to read/write. Allowing users to write their own documents can be problematic because:
  • They could fill up storage space.
  • We'd have to make documents unique per user. We don't want user A seeing user B's written doc.
• LLM prompt evaluation happens in parallel with processing the user's message. This means that if the message triggers LLM evaluation, a document may have already been written by that point.

[gsproston-scottlogic]

Adding question label as this may need splitting into smaller tickets. It's a good chunk of work.