Update resolvers/regex doc section (once a new release of `ua-parser-rs` has been cut)

[masklinn]

Cf

• Correctly rewrite bounded repetitions following an explicit class uap-rust#29 (significant reduction in memory consumption of the rust impl)
• Use an atom length of 2 for the regex filtering uap-rust#31/Re2 bench updates, revert #31 for UAs and OS uap-rust#32 (very significant perf improvement of the devices matcher with atoms of size 2)
• Investigate updating the re2 matcher to use an atom length of 2 #263 (can't tune atoms size in re2)

Meaning the memory overhead should be dramatically lower than it used to and competitive with re2, and the performance should be even better comparatively.

[masklinn]

Probably need to check against the old measurement method to make sure it's consistent, but here's uv run -m ua_parser bench --bases <parser> --caches none:

	time	time/line	maximum rss	peak memory
basic	35.51s	473us	48414720	16172544
re2	2.54s	34us	60030976	19007040
regex (current)	1.40s	19us	158285824	17565312
regex (next)	1.22s	16us	88113152	17057280

So:

• no clue what peak memory is in macos's time -l but it's useless
• regex.next has not quite caught up with re2 on memory, but goes from 2.6x (or +160%) to 1.46x (or +46%) memory overhead
• regex.next also improves runtime by 15% overall, which matches Use an atom length of 2 for the regex filtering uap-rust#31 with some additional overhead, and regex.next is now twice as fast as re2 on the bench (versus 80% faster previously)

Once regex.next is merged, basically the only reason to use re2 is "we already use re2".