[Enhancement] mobieusAI: pre-indexed snippet cache for the Q&A flagship

What

Today every mobieusAI question runs the FULLTEXT retrieval query against the source tables. Most communities have a long tail of common questions ("how do I change my password", "how do I make a forum private") that get asked over and over. Caching the top-K snippets per question hash for ~15 minutes cuts the average query path to a single Anthropic call.

Where it lives

• Surface: behind the Q&A drawer; invisible to users
• Backlog id: B-AI-RAG-3

Implementation notes

• Cache key: SHA-256 of (tenant_slug + scope + normalized question)
• Storage: existing FileCache service with 15-min TTL
• Cache stores the retrieved snippets only — the Anthropic call still happens per ask (answers can vary)
• Invalidation: any write to knowledge_pages / forum_threads / forum_posts clears the cache for that tenant
• Estimated win: ~40% reduction in DB read load for common asks

Tracked as B-AI-RAG-3 in docs/BACKLOG.md.