feat(mcp): add Hy3 data analyst server

[dujunjin]

Summary

• add a stdio MCP server for CSV, JSON, JSONL, and NDJSON analysis powered by Hy3
• expose three documented tools: profile_dataset, analyze_dataset, and generate_data_report
• add path confinement, file/row/sample limits, environment-only credentials, and API-key redaction
• include one-command uv installation, bilingual documentation, client configs, sample data, and a demo GIF

Why

This implements the data-analysis scenario from #3. MCP clients can inspect a dataset locally, ask Hy3 evidence-grounded questions, and generate a structured Markdown report without hard-coded credentials or public deployment.

Client integration

• WorkBuddy 2.115.0 bundled CodeBuddy CLI: loaded the server and called profile_dataset
• MCP Python SDK ClientSession: initialized over stdio, listed all three tools, and called profile_dataset
• project-level templates included for CodeBuddy, WorkBuddy, and Cursor

The sample CSV returned 6 scanned rows, 6 columns, and one missing value in units in both client checks.

Validation

• uv run --no-sync ruff check .
• uv run --no-sync pytest — 17 passed
• uv run --no-sync python scripts/check_mcp.py
• source distribution and wheel build successfully
• all client JSON templates parse successfully

The local validation environment did not provide a reachable Hy3 endpoint or API key. The Hy3 path is covered by mocked integration tests that verify the OpenAI-compatible request payload, reasoning mode, grounded dataset context, response handling, and client cleanup.

Install

cd mcp-servers/hy3-data-analyst
uv tool install .

Addresses #3.