Let `matchers::BuildError` implement `std::error::Error`

[oscargus]

Not sure if there is a particular reason that it doesn't? (Nor am I sure if I understand enough to grasp the consequences, no_std?, but see tokio-rs/tracing#3033 (review) for context.)

Anyway, to me it seems like this can be a good thing for more cases and may be easier than creating another error type?

[oscargus]

Ah, one should probably read the code first. I now realize that it comes from regex_automata::dfa::dense, so not the correct location for the issue.