Add safety intelligence schemas and refinery RAG knowledge base

[Ashish331-naik]

Summary

Adds the schema contracts and refinery-safety RAG foundation for SpecGuard.

Changes

• Added shared safety enums and validated Pydantic schemas
• Added contracts for shift logs, extracted safety events, historical incidents, risk-engine input, evidence, recommendations, and intelligence responses
• Added PDF, Markdown, and text document ingestion
• Added conservative cleaning and deterministic chunking
• Added safety metadata enrichment
• Added local embedding and persistent vector retrieval support
• Added six synthetic refinery SOPs
• Added authorized refinery incident and safety documents
• Added manifest-based document management
• Added RAG CLI commands for ingestion and retrieval
• Added tests and beginner-friendly documentation

Validation

• 35 backend tests passed
• 18/18 documents loaded
• 72 chunks created and indexed
• Retrieval smoke test returned relevant:
  • hot-work guidance
  • gas-testing guidance
  • equipment-isolation guidance
  • evacuation guidance

Safety

• Synthetic SOPs are clearly labelled as prototypes
• Recommendations remain human-approved
• No paid LLM API is required for ingestion or retrieval

[Ashish331-naik]

@copilot resolve the merge conflicts in this pull request

[Ashish331-naik]

@copilot resolve the merge conflicts in this pull request