Use three-valued logic with missing values in ==(::Tuple, ::Tuple)

[nalimilan]

For consistency with AbstractArray (missed by #24653).

Also add a test for the special case when the result of equality test is not missing even in the presence of missing values.

[nalimilan]

@nanosoldier runbenchmarks(ALL, vs=":master")

[ararslan]

I could be missing something super obvious, but how is this different than the test a few lines down that does !=?

[nalimilan]

Yeah, it's the same since the fallback is defined using ==, but that's the assumption we're checking. :-)

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[nalimilan]

OK, the "simd/local_arrays" regression cannot possibly be related. The "tuple/reduction" benchmark doesn't use == either. The "dates" regressions are also noisy benchmarks -- and so are the improvements. So overall I think we can conclude that nothing changed. Anyway if there was a type unstability, it would have a really dramatic impact.

[nalimilan]

I'll merge tomorrow barring objections.

[StefanKarpinski]

LGTM: the scheme is that isequal always returns a boolean and == uses 3VL, right?

[nalimilan]

Yes. BTW, I've noticed < and > need a similar change (with isless always returning Bool like isequal).

[StefanKarpinski]

Circle CI failures look legit and related, btw.

[nalimilan]

Circle CI failures look legit and related, btw.

No, AFAICT that's #24984. Let's see what a new run says.

[nalimilan]

CircleCI still fails, but with ProcessExitedException. Merging.