[cbaecker]

Follow up on:

• feat: Improve performance of built-in-functions via static regexes #249
• feat: Do not use regex for simple pattern-replacements to improve performance #252

Adds a fastpath for KeyMatch2()/KeyMatch3(), checking if regex-evaluation is actually required.
My test with 1000 random (failing) requests resulted in a ~6 times faster performance.
Real-world scenario on embedded was about 2 times faster for initial (non-cached) request.

[hsluoyz]

Thanks @cbaecker, the fast path idea is good and the numbers made the case. I've pushed a reworked version directly to master in 2e99457, so I'm closing this PR in favor of it:

• The fast path is a single find_first_of over the regex special characters (plus : for keyMatch2). With no special character it's plain equality; with only a trailing /* it's a prefix check. Everything else still goes to the regex.
• The trailing /* case also checks that the rest of key1 has no \r or \n, because . in /.* doesn't match line terminators. This keeps it exactly equivalent to the regex.
• keyMatch4 uses the same fast path, and keyMatch5 gets it through keyMatch3.
• Compiled regexes are cached per thread (thread_local, capped at 1024 patterns), so patterns with placeholders and regexMatch no longer recompile on every call either.

I checked it against the old regex-only keyMatch2/keyMatch3 on ~520k random key/pattern pairs with no differences, and added a few test cases. On a small policy set of plain, /* and :id paths it's about 10x faster.