refactor: let the schema decide an option's default

[mcanouil]

_schema.yml declares a default for each of the seven options, and a table in the Lua declared the same seven again. The schema check already answers what the schema says, and that answer was thrown away, so the literals decided while the schema was what the editor showed. Changing a default in _schema.yml alone did nothing, and nothing reported the difference.

The check's answer is kept now, and the table is demoted to a fallback for the two cases the schema cannot answer: a format the extension does not act on, where the check never runs, and a schema that could not be read, which is a state the extension renders through rather than stopping for.

The schema is read first and the fallback fills only what it leaves unanswered. The other order would have made the fallback the list of options allowed to have a default at all, which is the coupling this removes.

Three checks cover it. The wrapper name comes from the schema, it follows a change made in the schema alone, and it falls back when the file is gone. The wrapper is the option under test because it reaches the Typst output word for word, and the middle check also requires the old name to be absent, so a rename that missed a call site fails.

The three hot-fix defaults are still kept by hand. They are nested in the schema and read on their own path, which takes a boolean or a map, so they need more than a table of defaults handed to the options reader. That is tracked separately, and the comment in the code says so rather than claiming the schema decides everything.

No behaviour changes, so there is no changelog entry. Nineteen tests pass, example.qmd renders byte-identical to main in HTML, Typst and Reveal.js, and the documentation site renders with no warnings. Three review rounds.