ARROW-12955: [C++] Add additional type support for if_else kernel

[nirandaperera]

This PR adds fixed and variable binary type support

[github-actions]

https://issues.apache.org/jira/browse/ARROW-12955

[lidavidm]

This looks good on a first pass. I suggested some minor simplifications.

[nirandaperera]

@pitrou @lidavidm I made some changes to the code. As Antoine pointed out previously, the null values in the output string array would have non-empty slots in the data buffer. I think it requires a small fix. I'll add that here as well. But I'd like to get your feedback on the code flow.

[lidavidm]

IMO these cases could be consolidated into one or two checks since there are only 12 cases here (or 5 if you want to prune: null cond, true cond with null/non-null left array, false cond with null/non-null right array).

[nirandaperera]

well, there are 64 cases TBH. Each param can be null/ non-null and scalar/ array (4 variations). Since we have 3 params we have 4x4x4 = 64 permutations. 🙂

[nirandaperera]

CheckWithDifferentShapes actually checks for all those cases + with offsets. But I agree, in CheckWithDifferentShapes it could check similar cases redundantly in several iterations. But we decided to leave it as it is.

[lidavidm]

minor nit but I wonder if the compiler could generate similar (enough) code from a

struct CopyBinaryBulk {
    const offset_type* offsets;
    const uint8_t* values;
    offset_type* out_offsets;
    uint8_t* out_values;

    operator()(...) { ... }
};

so you don't have to copy-paste these everywhere

[lidavidm]

N.B. it looks like this segfaults in the tests on Windows MSVC 2019 (https://github.com/apache/arrow/pull/10538/checks?check_run_id=2966481230#step:8:339) and there are various errors on Clang, MSVC, and MinGW (mostly related to implicit casts or needing to explicitly parameterize std::max).

[lidavidm]

nit: just data() here?

[lidavidm]

maybe checked_cast<const FixedSizeBinaryType&>(*left.type()) or checked_pointer_cast<FixedSizeBinaryType>(left.type()) (will be dynamic cast in debug, static cast in release)

[lidavidm]

Looks like this also needs to be rebased now. But once that's done and the nits above are addressed I think this can be merged.