Add multiformat-hybrid-rag agent sample

[lspataroG]

Summary

• Adds a production RAG agent with multi-format document ingestion (PDF, Office, HTML, JSON), contextual chunk enrichment, and hybrid semantic + keyword search via Vector Search 2.0
• Serves through an ADK agent, REST API, and MCP server from a single Cloud Run service
• Includes full ingestion pipeline on Vertex AI Pipelines, Terraform IaC, and Agent Starter Pack integration

Key features

• Multi-format parsing: PDF (Gemini multimodal), DOC/DOCX/PPT/PPTX (LibreOffice), HTML, JSON/JSONL, Markdown
• Contextual chunking: Gemini-generated per-chunk context prepended before embedding
• Hybrid search: semantic + BM25 via Reciprocal Rank Fusion on Vector Search 2.0
• Three serving interfaces: ADK agent chat, REST search API, MCP server
• Full ingestion pipeline: 3-step Vertex AI Pipeline (preprocess → chunk & index → cleanup) with daily scheduling
• Infrastructure as code: Terraform for BQ, VS2, Cloud Run, IAM
• Agent Starter Pack: agent-starter-pack create my_app -a adk@multiformat-hybrid-rag

Test plan

• make install succeeds
• make setup-infra provisions all resources
• make data-ingestion-remote runs the pipeline end-to-end
• make local-backend starts the agent + REST + MCP server
• make playground launches ADK playground UI
• make deploy deploys to Cloud Run
• Search API returns grounded answers from indexed documents
• MCP tool ask_knowledge_base works from external clients