markdown: parse fenced_code_attributes extension

[edwintorok]

Pandoc supports this extension: https://pandoc.org/MANUAL.html#extension-fenced_code_attributes

And this:

Recognize them in the lexer. Try to limit the complexity of the regular expression by splitting off parsing of attributes into a separate 'parse' (otherwise we hit automata size limits in ocamllex).

According to
https://quarto.org/docs/authoring/markdown-basics.html#ordering-of-attributes the ordering has to be:

• #identifiers
• .classes
• key-value attributes

For now on output we always normalize to this form (which isn't ideal, but could be improved later):

I initially tried to fully parse the attributes, but I've exceeded the maximum size of the ocamllex automaton, so I kept it simple in this PR (and do some minimal parsing in OCaml later, note that key-value pairs aren't split correctly, but when joined backed together they retain the original value).

[edwintorok]

(this PR might need some wider testing to check it doesn't break backwards compatibility, is there a larger corpus you'd normally test changes like this on? e.g. the realworldocaml book, anything else?)