ARROW-12835: [C++][Python][R] Implement case-insensitive match using RE2

[lidavidm]

This uses RE2 to implement a case-insensitive substring search.

Originally, I implemented this using utf8proc, but then found it was about an order of magnitude slower than RE2. (This isn't an apples-to-apples comparison; utf8proc does it more 'properly' and handles more Unicode corners.) So I switched to just doing it with RE2 instead, especially since the utf8proc approach was complicated. (You can still see it in the original commit here if you're curious.)

[ianmcook]

Bikeshedding here, but maybe we should rename the C++ and Python argument to ignore_case instead of case_insensitive which is a little easier to understand without getting mixed up in double negatives

[github-actions]

https://issues.apache.org/jira/browse/ARROW-12835

[lidavidm]

ignore_case is also what Python's regex engine calls it at least (though not RE2), so I've renamed it.

[pitrou]

Could you give an example where re2 doesn't follow proper Unicode semantics?

[pitrou]

I don't understand why two MatchSubstring and MatchSubstringImpl classes. It seems one should be sufficient?

[lidavidm]

There's only one of each. I moved them around in this PR, but it's the same as before.

[lidavidm]

An example of RE2's Unicode handling is here: google/re2#262

That said I don't think it's too big a deal for us.

[lidavidm]

There's only one of each. I moved them around in this PR, but it's the same as before.

[pitrou]

That said I don't think it's too big a deal for us.

It depends what you mean. The fact that ß and ss don't match is a bit of a bummer for German text, for example. I don't know what the intended use case is.

[lidavidm]

That said I don't think it's too big a deal for us.

It depends what you mean. The fact that ß and ss don't match is a bit of a bummer for German text, for example. I don't know what the intended use case is.

It is a bit of a bummer but I think it's also not 'unexpected' in that other systems (languages, etc) probably make the same tradeoff, hence why I thought it was worth a note in the docs, but wasn't worth implementing a full solution using utf8proc.

[pitrou]

Rebased to check for CI.