simplify jflex grammars by using difference rather than negation

[rmuir]

This change uses a new jflex feature (jflex-de/jflex#654) to simplify emoji processing in the grammar. We can do a set difference rather than workaround it with complement + demorgan stuff.

It is cosmetic: doesn't change the resulting tokenizers (see diff), but makes the emoji parts easier to read.

bonus: major speed up to regenerating that huge UAX29UrlEmail DFA.

Before:

> Task :lucene:analysis:common:generateUAX29URLEmailTokenizer
Aggregate task times (possibly running in parallel!):
 918.87 sec.  generateUAX29URLEmailTokenizerInternal

After:

> Task :lucene:analysis:common:generateUAX29URLEmailTokenizer
Aggregate task times (possibly running in parallel!):
 285.26 sec.  generateUAX29URLEmailTokenizerInternal

This was suggested by jflex developers to help with the very-slow-regeneration on jflex-de/jflex#715 . It doesn't solve all of our problems there, but it makes things a lot less painful :)

[rmuir]

I'll take care of the precommit. There's some build wierdness in the regeneration where we pull the TLDs and make the new included-TLD.jflex after we recompile the grammar. Maybe dependencies are backwards. Anyay, I didn't really want to suck in today's new TLDs and trigger any changes to the parsers anyway, as I wanted to show this change "makes no difference".

Oh well, ill just force-regenerate everything (you really need to do it twice right now to really pull in the new TLDs) and push again.

[rmuir]

and don't worry about the gradle build, I think part of the issue is that this TLDs file annoyingly changed while I was iterating and working on this:

# Version 2021120400, Last Updated Sat Dec  4 07:07:01 2021 UTC

[rmuir]

I opened https://issues.apache.org/jira/browse/LUCENE-10285 about the TLD task dependency. For now, I just ran regenerate again and it brought in the TLD changes. It is a lot less painful at least now that it takes 1/3 of the time!

[dweiss]

That's a nice improvement!

[rmuir]

@dweiss i think it's good win just for getting a simpler grammar.

There is probably evil stuff we could do to speed up the monster. Yes, I am slightly tempted to import org.apache.lucene.util.automaton into GenerateJFlexTLDMacros.java... but I agree with your thoughts on the JFlex issue, it is better to just generate the simple "transparent" grammar of TLDs, despite how slow it makes things.