feat(mcp): add Hy3 LeadIntel MCP server

[Hanlin-YE]

Summary

• Add mcp_servers/leadintel, a Hy3-powered local stdio MCP server for B2B lead intelligence and sales/research operations.
• Expose 5 MCP tools: analyze_lead, query_knowledge_base, generate_outreach_plan, batch_score_leads, and hy3_leadintel_status.
• Include CodeBuddy/WorkBuddy, Cursor, and Cline MCP config examples, sample knowledge-base docs, sample leads, an offline demo transcript, and a GIF.

Why this scenario

This project targets a distinct gap from existing generic RAG, CSV analysis, code review, and deep research MCP submissions: a vertical B2B workflow that connects lead scoring, grounded product evidence, outreach planning, and batch lead operations.

Validation

• python3 -m pytest -q -> 5 passed
• PYTHONPATH=src python3 -m hy3_leadintel_mcp.server --selfcheck -> PASS
• python3 scripts/sdk_stdio_client.py -> MCP SDK stdio initialize, tools/list, and tool calls succeeded; 5 tools listed

Notes

• Offline mode is deterministic when HY3_API_KEY is absent, so reviewers can verify installation and MCP protocol behavior without a key.
• Real Hy3 mode uses an OpenAI-compatible /v1/chat/completions endpoint via HY3_API_BASE, HY3_API_KEY, and HY3_MODEL environment variables.
• API keys are never hardcoded; status only reports api_key_present as a boolean.

Related to #3.

[Hanlin-YE]

Update: added real MCP client verification evidence for the hard requirement of 2 different MCP clients. Verified Claude Code CLI 2.1.146 connected and called hy3_leadintel_status; verified MCP Inspector CLI 1.0.0 listed 5 tools and called hy3_leadintel_status with isError: false. Evidence is documented in mcp_servers/leadintel/assets/client_verification.md, and assets/demo.gif now reflects the two-client verification flow.