[Feature Request] Permission-aware semantic retrieval API for the LLM index (+ configurable chat limits)

Use case I run Paperless-ngx at home as the archive for my family: insurance policies, medical lab results, warranties, house renovation documents, notes about our heating system. My wife and kids have their own Paperless users with restricted permissions. On top of Paperless I'm building a small assistant that my family can ask questions through a Telegram bot ("when does the house insurance expire?", "has my ferritin gone up or down over the last three years?"). The assistant is an LLM agent that talks to Paperless through the REST API (via an MCP server) using each family member's own API token, so Paperless permissions stay the single source of truth. Keyword search works, but our documents are in Lithuanian, which has heavy inflection, so the agent often misses relevant documents. The LLM index that Paperless 3.x builds already solves this, but right now it can only be used through the built-in chat. What I'm asking for 1. A retrieval-only …