Extract queue from Knowledge base

[sorenlouv]

Background
When adding pre-loaded knowledge base entries (like Lens docs) we need to wait for the Knowledge Base (aka Elser model) to be available before writing and make sure that only a single Kibana instance ingests the documents in order to avoid spamming the cluster with duplicate entries.

We currently use the task manager and a custom queue to handles this. The current implementation adds complexity to the knowledge base, making it harder to make changes and write tests for this. I suggest we extract the queuing logic from the knowledge base. If needed we can create a completely separate queuing service or better look into getting rid of the queue completely.

The queue is implemented in the following services:

• Obs AI Assistant Service
• Knowledge base service

Acceptance criteria

• Pre-loaded entries should await ingesting until knowledge base is ready (queue or no queue?)
• Pre-loaded entries should only be ingested by a single kibana instance. (the task manager can handle this for us)
• Long entries should be chunked (split into smaller docs). This will (eventually) be handled automatically by semantic_text for both pre-loaded and user created entries
• The API should throw an error if a client attempts to add an entry before knowledge base is ready (status quo).

Depends on: #186499

[sorenlouv]

The queue was removed in #191043