Generate extension docs from raw metadata

[makhnatkin]

What we want

• Continue the extension docs pipeline explored in ci: add extension docs generation pipeline #1067, but land it incrementally.
• Start with deterministic raw extraction: category, schema, actions, keymaps, plugins, options, examples, and presets.
• Use raw metadata as the base for generated Markdown docs and later enrichment/assembly steps.
• Make every generated artifact reviewable and easy to extend.

Why

• PR ci: add extension docs generation pipeline #1067 shows the full direction, but the pipeline is easier to review and stabilize in smaller parts.
• Manual docs drift from real extension APIs and behavior.
• A raw IR gives us a stable contract before adding AI-assisted or richer documentation layers.

Principles

• Source code is the source of truth.
• Prefer deterministic extraction first; keep enrichment explicit and optional.
• Extract only static, confident data; avoid guessing.
• Keep raw data separate from presentation.
• Keep docs-gen isolated in infra/docs-gen and covered by focused tests.