Previously, all possible parse instances of a string were shown, which
included the unquoted, single quoted, and double quoted versions,
meaning that "always" was shown as:

    string(([ws] & ('always'
		    | ("'" & ['a', 'l', 'w', 'a', 'y', 's'] & "'")
		    | ('"' & [rich_char(c='a'), rich_char(c='l'), rich_char(c='w'), rich_char(c='a'), rich_char(c='y'), rich_char(c='s')] & '"'))))

This changes the output to:

    string([frequency])

Extend DocStringTag to include "all" known types, along with document
how each of them behaves. This also forced a few changes to its
behaviour, which is reflected in the updated tests. The biggest change
is that we now barely test for valid "tag_name" values, since a valid
tag should always have a valid (albeit unknown) tag name.

Errors are now created through a factory, embedding information about
WHERE the error comes from in the source.

Also add a clearer indicator to the error output.

As a steping stone, the raw match data was emited to the frontend. This
just properly renders it.

If `parse_puppet` crashed, instead format the the raw code, adding an
error marker on the erroring character if possible. The error message is
also added.

These will later be used for arrows pointing at errors.

A match literal was already literally a string. Instead, tag matches
with data for clearer information.

Previously the the input json was directly mapped into the output
strings, through different formatters. This changes it so we instead
first build a proper AST of python objects, which is later mapped to
output strings (still through formatters).

This gives us much better type information, and will give way to much
more advanced formatters in the future.

This includes some regressions:
* Docstrings aren't handled as well
* The code lacks any form of hyperlinks.

Misc includes:
    - at least one more category
    - basic sidebars

A simple normalization making it easier later.