Extend the scanner to references, paragraphs, media, vocabularies and menus

Problem/Motivation

A flat list of bundles and fields is not enough to plan an API. What decides the shape of a response is how bundles point at each other: which fields nest paragraphs, which reference media, which reference terms of a vocabulary, and what the menus look like — the last one because breadcrumbs and the static data resource are built from them.

Proposed resolution

• Add to the map: the reference graph between bundles, paragraph types with their own fields, media types with their source field, vocabularies with their hierarchy depth, menus with their tree.
• Detect reference cycles and record them — a bundle that references itself directly or through a paragraph is normal content modelling and a stack overflow at render time if nobody noticed.
• Record depth for nested paragraphs, since the generator needs a bound.
• Keep the optional dependencies optional: no paragraphs module, no paragraph section in the map, and no error.

Remaining tasks

Everything.