Fix scalar eq for booleans

[gatesn]

And derive partial eq where we can

[robert3005]

What was broken here? Hard to figure out from the diff

[joseph-isaacs]

We want scalar to only be comparable to other scalars of the same dtype

[robert3005]

I thought top level scalar did the cast. But I guess you can still get a typed view

[gatesn]

This check makes sure they have the exact same nullability

[gatesn]

This is raising some bigger questions:

• When we convert scalars from DataFusion, we do a really bad thing... which is return Scalar::Null for null literals...

  vortex/vortex-scalar/src/datafusion.rs

  Line 154 in d0294e4

  .unwrap_or_else(|| Scalar::null(DType::Null))

• Vortex has chosen to never implicitly cast... but should we relax this rule for nullability only? If we do, then our conversion from datafusion should be as tight as possible and then upcast where necessary.

[robert3005]

I think we can fix up the conversion to be scalar null of dtype?