Ship personalised RAG, 10m SLA, parallel generate, and AI phase docs for HF pilot.

feat: RAG upload sanitisation layer before vector indexing

feat: async pipeline, job queue, generation hardening, and docs

feat: adaptive writing style — user docs first, KB fallback, few-shot examples

- Added support for RICS survey levels (1, 2, 3) in document uploads and reports, allowing for better tier management and retrieval filtering.

Added canonical rollout functionality, including new API endpoints and request/response models for upgrading indexed content. Updated main application to include the canonical rollout router. Introduced a new document loading function for raw documents in the ingestion pipeline. This enhances content management and semantic upgrades across the application.

- Replaced timezone.utc with UTC from datetime for consistent datetime handling.

- Introduced `ingest_timeout_seconds` setting in config.py to define maximum time for document ingestion.

Merge origin/experiment-lab (LangChain RAG) into experiment-lab + product features

- Updated `.env.example` to include new settings for bulk upload limits and batch processing.

feat: enhance document upload and processing capabilities with new personalisation features

feat(experiment-lab): rewrite RAG pipeline using LangChain framework

fix: apply all bug fixes and improvements (embeddings, cache, postprocess, frontend, tests)

feat: scaffold complete RAG system for RICS-style report generation