Fixes the most recent issue observed where LaTeX was not being rendered,
despite apparently correct delimiters.

The issue came from competing delimiter renderers. Previously we were
modeling each set of acceptable delimiters (e.g., `$$` and `$`)
independently. This was apparently causing the tokenizer to get confused
in certain cases.

Fixes by unifying the delimiters in a single extension. So now we
consider the best match for a delimiter in a given span of text among
all candidates (i.e., the "longest first" delimiter encountered) and
render accordingly.