What this PR does / why we need it:

Readme at https://github.com/ntkathole/feast/tree/agent_blog/examples/agent_feature_store
Blog at https://github.com/ntkathole/feast/blob/agent_blog/infra/website/docs/blog/feast-agents-mcp.md

Adds a complete working example and blog post demonstrating how Feast serves as both a
context provider and persistent memory layer for AI agents via the Model Context
Protocol (MCP).

examples/agent_feature_store/ — End-to-end agent example with:

• Four Feast-backed tools: lookup_customer, search_knowledge_base, recall_memory, save_memory
• Agent memory as a first-class Feast feature view (agent_memory) — entity-keyed, TTL-managed, governed by RBAC
• Role-based responses (same question yields different answers for Enterprise vs Starter customers)
• Memory continuity across conversations (returning customer is recognised without re-explaining)
• Demo mode (no API key needed) and live mode (OpenAI, Ollama, or any OpenAI-compatible endpoint)
• One-command setup via ./run_demo.sh

infra/website/docs/blog/feast-agents-mcp.md — Blog post:
"Building AI Agents with Feast: Feature Stores as Context and Memory"

• Explains the problem (agents need context, memory, and governance)
• Shows how Feast + MCP turns a feature store into a governed agent tool
• Walks through the customer-support agent example
• Covers production architecture (MCP Gateway, sandboxed containers, observability, Kagenti)