Improve handling of <Listing>s

[chriskrycho]

• Add support for listings which have file names but no caption or number. This means that we now support having all such be optional; in that case we still emit some nice semantic HTML with the listing wrapped in a <figure>, just without a <figcaption> or a <span> for the file name.

• Add support for having captions like "This is a caption about a Box<T>". XML parsers reject this, because XML does not allow the < or > to appear in attributes unescaped. We could stick with xmlparser, and attempt to switch to using owned strings which we preprocess to escape them, but that is a tricky dance: we also need to run Markdown handling on them, so we would somehow need to do so prior to parsing the XML at all, so that we do not end up with the literal > appearing in the rendered output.

  This PR sidesteps all of that complexity and resolves the underlying issue by switching mdbook-trpl-listing over to html_parser, which is a much lighter-weight crate which simply implements a compliant HTML parser using Pest. I read the source today and it is quite reasonable, and is much smaller and simpler than xmlparser. This works because HTML, unlike XML, is totally fine with having < and > appear within the string bodies of attributes. I also evaluated html5ever, but while that is notionally more robust, it also comes with substantially more wiring to do and a harder-to-use API.

These two changes should unblock the rest of the work on #3919.