Shared CPU ML base image for Data Vault AI services.
This image owns the CPU embedding stack used by rag-agent, sql-agent, and
vectorization-service-cpu, including CPU-only PyTorch, sentence-transformers,
Transformers, ChromaDB, and LangChain integration packages. Centralizing those
dependencies here keeps service rebuilds fast and avoids version drift such as
regex / transformers mismatches.
GPU vectorization must not use this image because it needs a CUDA-capable PyTorch runtime.