Document paragraphs

[Horusiath]

TODOs:

• 1. Split documents by paragraphs, not arbitrary length (see: chore: replace document plain text with paragraphs AppFlowy-Collab#361)
• 2. Create af_collab_embeddings.fragment_id using a consistent hash of chunk it relates to.
• 3. At the beginning of indexing, get collab content with list of its currently indexed fragment_ids.
• 4. When generating embeddings skip embeddings for those chunks which fragment_id was found in the point above and keep their content unchanged.
• 5. When persisting embeddings don't delete all of the existing ones. Delete only those that were not found in point 3, those for which we didn't got embeddings stay the same, and insert all of the new ones.