Expand description
Real TOML front-matter + markdown body parsing (plan Phase 2).
A corpus document is a +++-fenced TOML header followed by a markdown body:
+++
id = "D2"
kind = "decision"
...
+++
markdown body…The header maps to a Record (kind → Facet, with field-level errors collected
and joined into one ParseError); the body is scanned with pulldown-cmark for
bare-id mentions (skipping code), § section refs, links, and finding anchors.