Add regex error source to RegexError

[ratmice]

Marking as a draft, because this is perhaps questionable.

Here is the example source producing the error.

%grmtools {
  !unicode
}
%%
. "dot"

There are two questionable things here, the first being the Eq implementation relying on regex::Error's to_string() impl,
because regex::Error is only PartialEq.

The second being when we run the example through nimbleparse we see the regex::Error includes lrlex's wrapping of the regular expression e.g. \A(:? unfortunately, regex doesn't really give us any kind of Span we otherwise might be able to use to translate this back to the lex source locations.

Previously this just produced "Invalid regular expression", with this change it now produces a reason.
So, I'm a bit torn on this because I do feel like the extra information is helpful.

[Error] in bad.l
    5| . "dot"
        Invalid regular expression: regex parse error:
        \A(?:.)
             ^
    error: pattern can match invalid UTF-8

[ltratt]

This might be a dumb question but do we need PartialEq on LexErrorKind? I can't, off the top of my head, immediately think why it's there.

[ratmice]

It looks like it is used for merging spans of duplicate occurrences so that you get an error at a bunch of places,
rather than a bunch of errors each having a place, I think I added it, and for that reason.

[ltratt]

Aha, that makes sense. On reflection, I wonder if it would be better to make this a "normal" function called something like is_mergeable_with(&self, other: Self)? It gives us a "nicely weaker" external API (since we no longer promise PartialEq to other users).

[ratmice]

It was also used in the testsuite, so I went with the more generic is_same_kind().

[ltratt]

Please squash (and rebase to remove the conflict).

[ratmice]

Squashed.