ARROW-12843: [C++][R] Implement is_inf kernel

[lidavidm]

No description provided.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-12843

[ianmcook]

Thanks for doing this @lidavidm! You mind if I push a commit to your fork to tweak some of the R stuff?

[lidavidm]

Yeah, please go ahead! I admit I'm not familiar with the R side of things and mostly just pattern match on similar code.

[ianmcook]

The fact that is_inf returns false for NaN is inconvenient because its negation returns true.

Both R and NumPy have separate functions to test for finite and invite values, and these all return false for NaN:

> is.infinite(NaN)
[1] FALSE
> is.finite(NaN)
[1] FALSE

>>> np.isinf(np.nan)
False
>>> np.isfinite(np.nan)
False

I wonder if this suggests we should also make an is_fin or is_finite kernel to avoid this problem of negation.

[lidavidm]

Oh, doh. Yes, then having an is_finite kernel would make more sense than just inverting it. I'll update that.

[ianmcook]

This also fixes an inconsistency with the is.na() method for Scalar objects in the R package. Previously that returned a length-1 R logical vector. Now it returns a Scalar having the bool data type. This was necessary go make is.finite() and is.infinite() work consistently.

[cyb70289]

LGTM, +1

[cyb70289]

@ianmcook , do you have other comments? Can we merge this PR?

[ianmcook]

@cyb70289 No more comments from me, looks good, thanks