Support `bytes.matches(string)` and `matches(bytes, string)`

[rodaine]

CEL supports string.matches(string) to apply an RE2 regular expression to strings, but RE2 patterns can safely be applied to a non-UTF8 bytes string (the pattern itself must be UTF-8, however). Can we update the spec for matches to include this overload?

Context from protovalidate: bufbuild/protovalidate#268

[TristonianJones]

I'd be okay with adding this functionality, though it'd have to be part of one of the extension functions perhaps. What do you think?

[rodaine]

I think that's totally reasonable, mainly wanted to preserve the symmetry with between string and bytes types where it makes sense. I don't believe we had any users of protovalidate running into this (besides myself) so there's not a burning desire for it. And since it's more general than a protovalidate feature, figured it'd be better at the CEL level than a custom function in PV.

[TristonianJones]

Fair enough. I appreciate the willingness to get this surfaced as part of the supported CEL surface area. I think we're going to expand what gets included in 'cel-spec' to include canonical extensions and best practices for data modeling w.r.t. certain policy domains. So, I think there's room for your suggestion to surface in this repo as well.