ARROW-13130: [C++] Add decimal support to arithmetic kernels

[lidavidm]

This adds decimal support for the following kernels (and _checked variants where applicable): abs, acos, add/sub/mul/div, asin, atan, atan2, ceil, cos, floor, hash_stddev, hash_tdigest, hash_variance, is_finite, is_inf, is_nan, ln, log1p, log2, logb, mode, negate, power, quantile, round, round_to_multiple, sign, sin, stddev/variance, tan, tdigest, trunc

Most kernels cast decimals to double and proceed. Some, including rounding, directly operate on decimals. Aggregate kernels directly operate on decimals (and cast to double inline) since DispatchBest is not usable for the aggregate nodes (at least, unless we also reimplement implicit casting there).

Additionally, ValidateFull for scalars/arrays now checks FitsInPrecision. A number of tests were adjusted to account for this.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-13130

[lidavidm]

Apologies in advance to whoever tries to review this. If it helps, I can try to split along these lines:

• Non-kernel changes (changes to ValidateFull and fixing related tests),
• Arithmetic kernels except rounding,
• Rounding kernels

[westonpace]

This can be a follow-up but can we modify compute.rst to flag which operations cast decimal to double? Probably as a new flag.

[lidavidm]

This can be a follow-up but can we modify compute.rst to flag which operations cast decimal to double? Probably as a new flag.

D'oh, I knew I was forgetting something - I'll update the docs. I'll also split out the array validation changes and the rounding kernel here (since that kernel is much more involved and honestly could probably use some cleanup).

[cyb70289]

Thanks @lidavidm, this is a lot of work to add decimal support to all these kernels.
I didn't review hash aggregate related code, all other changes LGTM.

[lidavidm]

@pitrou would you be able to take a look at the hash aggregate changes here?

[pitrou]

+1, thank you

[ursabot]

Benchmark runs are scheduled for baseline = 3626a08 and contender = 6f4c991. 6f4c991 is a master commit associated with this PR. Results will be available as each benchmark for each run completes.
Conbench compare runs links:
[Finished ⬇️0.0% ⬆️0.0%] ec2-t3-xlarge-us-east-2
[Finished ⬇️0.0% ⬆️0.0%] ursa-i9-9960x
[Finished ⬇️0.49% ⬆️0.54%] ursa-thinkcentre-m75q
Supported benchmarks:
ursa-i9-9960x: langs = Python, R, JavaScript
ursa-thinkcentre-m75q: langs = C++, Java
ec2-t3-xlarge-us-east-2: cloud = True