There is no need to allocate the capture group vectors in the matcher state when the only capture group is 0, because capture group 0 is just the whole match and we already know the bounds of the whole match (start is the position where _Matcher2::_Match was called, end is the value of _Cur in the matcher state).

I can't come up with any additional test cases that should be added for this change and I think the existing test coverage should already catch it if I made a mistake here.

Benchmark on my machine: