Add schemas for all JSON extracts

[tidoust]

This provides a first level of schema validation for curated data extracts, see #657 for context.

Goal is to make it easier to detect and document (through a changelog, so also useful for #704) situations where we change the structure of data extracts.

Schemas, notably those that deal with parsed IDL structures, could go deeper into details.

Tests are run against the curated version of data. That is not necessary for extracts that aren't actually curated (dfns, headings, ids, links, refs), just more convenient not to have branching logic in the test code.

Creating the PR as pull request as 69 of the new tests currently fail, either because extraction logic in Reffy needs to be slightly improved to create more consistent data structures, or because of actual issues in the specs themselves (e.g. invalid URL fragments).