Implement Logicals

[andy-thomason]

This PR does the following:

Add Rbool as a type - Bool is still there but will be deprecated.
Add Logicals as a type.
Redesign the gen_vector_wrapper_impl macro a little, but keeping Ilia's original design.
Make sure everything works with scalar types including Integers and Doubles.
Implement ALTREP logicals support.
Add Logicals to Rany.
Fix many many tests.

[Ilia-Kosenkov]

What about a few trivial tests on R side here https://github.com/extendr/extendr/blob/master/tests/extendrtests/src/rust/src/lib.rs?
E.g., test for NA and elt access, similar to Integer and Doubles?

[andy-thomason]

@clauswilke Tell me if I should drop Bool in this PR?

[clauswilke]

I think dropping Bool now is fine. The current version of extendr is basically a complete rewrite of the previous version anyways, so I don't see any reason to keep backwards compatibility.

[andy-thomason]

Good. I'll do this then.

[andy-thomason]

We may need to review the current case that:

NA_LOGICAL != NA_LOGICAL

Because PartialEq never returns true for NAs on Rbool, Rint or Rfloat.

[Ilia-Kosenkov]

We may need to review the current case that:

NA_LOGICAL != NA_LOGICAL

Because PartialEq never returns true for NAs on Rbool, Rint or Rfloat.

Can you explain this? It seems that if PartialEq never returns true for NA, then NA_LOGICAL != NA_LOGICAL, or am I missing something?

[clauswilke]

On the R side, you cannot compare an NA to an NA. The result is neither TRUE nor FALSE but rather NA. So returning TRUE when comparing NA to NA would definitely be problematic.

[Ilia-Kosenkov]

I am aware of this feature of R. Why can't we treat all NAs as floating-point NaNs, i.e. any equality operation involving an NA results in false, even NA == NA. This is somewhat consistent with R implementation, as I doubt anyone treats logical NA obtained during comparison as TRUE.

[clauswilke]

I agree, that's the most reasonable choice. It does imply NA_LOGICAL != NA_LOGICAL, though. Maybe I misunderstood what you were trying to say earlier. I though you considered it a problem.

[Ilia-Kosenkov]

I was trying to say NA != NA is the approach I prefer. Sorry for confusion.