[kakkar87d]

email and idn-email format regexes backtrack quadratically on long invalid strings

>>> v = fastjsonschema.compile({'format': 'idn-email'})
>>> v('a@a' + '.' * 80000 + '@')    # master: 61 s, this branch: 1 ms
>>> v = fastjsonschema.compile({'format': 'email'})
>>> v('a@' + 'a.' * 40000 + '@')    # master: 24 s, this branch: 1 ms

input length	email master	email here	idn-email master	idn-email here
10 KB	0.27 s	0.1 ms	0.53 s	0.1 ms
40 KB	4.8 s	0.4 ms	10.1 s	0.5 ms
80 KB	23.8 s	0.8 ms	61.4 s	1.0 ms

After the @, [^@]+\.[^@]+\Z retries every dot once the trailing @ makes \Z fail, and the email lookahead (?!.*\.\..*@) rescans the rest of the string for each ... Format checks run on the data being validated, so a single 80 KB string field in a request body keeps the validator (and the GIL) busy for about a minute.

The local part is now matched as dot-separated runs of [^@.], and the "dot inside the domain" requirement is a single lookahead, (?=[^@][^@.]*\.[^@]), so there are no overlapping choices left to retry. Comparing old and new regexes over every string up to length 10 built from a, ., @, \n (plus 300k random strings), idn-email accepts exactly the same strings. email differs only when a newline sits between a .. in the local part and the @ (e.g. 'a..\nb@c.d'): the old lookahead's . could not see past the newline and accepted it, the new regex rejects it.

tests/test_format.py gets email / idn-email cases and a long-input test that fails on master (about 45 s) and passes here. The full suite passes on Python 3.10 and 3.12, and pylint output is unchanged. CHANGELOG.txt has an entry under UNRELEASED.