`<regex>`: Consider special-casing some simple loop structures

Currently, loops like a* or \d* go through the whole NFA state transition processing for each repetition, which slows down the handling of such loops a lot.

I think it's probably worth it to recognize and special-case such simple and comonly used loop structures in the matcher. (But we should not add such special-case handling to non-simple loops and instead resolve #5957 to effectively extend the simple loop handling to branchless loops.)

Begin with the regex matcher and its NFA state-transition processing, the entry points named in the issue. Compare simple loops such as a* and \d* with non-simple loops; done means special-casing only the simple structures while leaving non-simple loops to the approach described alongside #5957.