Mimicks `a2x`'s behavior with regards to images, copying them into the
destionation repository. `a2x` parses the generated html files but we
don't want to use their parser because we'd like to make a clean break
from `a2x`. We can't use any of perl's wonderful parsers because we
don't want to add any additional dependencies to the project. Instead,
we use the AST of the parsed asciidoc files and look for images. This
isn't *quite* the same thing because it'll miss any images that we add
using as part of the xsl transformation but we can handle those
explictly in the perl code if there are any. In addition to not
requiring any more dependencies it has the advantage of build "built in"
to the asciidoctor work so we don't need to work on it later if we
change the perl orchestration stuff. Finally, it is pretty quick because
we don't have to parse all of the generated html - we can just use the
already parsed asciidoctor!