feat(mcp): package gknowledge-mcp for PyPI and the MCP registry

[francescobianco]

Refs #22. First half of the issue: everything needed to publish gknowledge-mcp on PyPI and in the official MCP registry. The community lists (awesome-mcp-servers, Glama) come after the package is live.

What's in it

• build-aux/pypi/: a separate gknowledge-mcp distribution with only the headless core and the MCP server. No GTK, no dependencies, so uvx gknowledge-mcp works on any OS. It has its own PyPI README with <!-- mcp-name: io.github.openapi/gknowledge -->, so the main README stays as it is. Build with make pypi.
• server.json (schema 2025-12-11, validated) for io.github.openapi/gknowledge, PyPI package, uvx runtime, stdio.
• glama.json to claim the Glama listing for the org repo.
• .github/workflows/mcp.yml: on v* tags or run by hand, it builds, publishes to PyPI through trusted publishing (skip-existing), syncs the server.json version with gknowledge/__init__.py, and publishes to the registry with GitHub OIDC.
• gknowledge-mcp --init PROJECT creates the project if it doesn't exist yet, so the server is usable without the desktop app. Without --init, nothing changes. Tests and man page updated.

Tested

• make test and pytest: 13 passed, including 2 new tests for --init.
• twine check passes on wheel and sdist. The wheel contains only gknowledge/{__init__,mcp}.py, core/ and providers/.
• In a clean venv without gi: initialize, tools/list, add_source and ask all work on a project created with --init.

One-time setup before the first run (maintainer)

1. On pypi.org, under Publishing, add a pending trusted publisher: project gknowledge-mcp, owner openapi, repo gknowledge, workflow mcp.yml, environment pypi.
2. Create the pypi environment in the repo settings (optionally limited to tags).
3. Run Actions → MCP package → Run workflow on main.

🤖 Generated with Claude Code