add a check for valtypename to comparedims to be used when broadcasting

[tiemvanderdeure]

Solves #878

E.g.

using DimensionalData, DimensionalData.Lookups
z = Z(1:2)
z2 = Z([1,2])
zcat = Z(Categorical(1:2))

rand(z) .* rand(z2) # still works
rand(z) .* rand(zcat) # now errors

With the error message ERROR: DimensionMismatch: Lookup types for Z do not match: Sampled and Categorical.

[rafaqz]

Isn't this too strong? Now Sampled and Projected will also error.

Which may not be what we want?

[tiemvanderdeure]

Okay probably not. So maybe we want to check if they have the same abstract type. So if both are AbstractSampled we should allow it. I'm not sure how to implement that robustly, though.

[rafaqz]

Yeah, this is probably why I didn't do it 😅

Like what does the same abstract type really mean when it's generally accepted that you can add intermediate abstract types without a breaking change.

[felixcremer]

We could introduce a function for this check and return false as fallback and return true for combinations that assume to be comparable

[rafaqz]

Still a bit complicated to know what the function should do.

We may need to just manually distinguish AbstractSampled from AbstractCategorical, and let promote_dims take care of the rest

[tiemvanderdeure]

Maybe this function could return true for identical types, or for combinations where there is some specialized implementation of promote_first (which is I think just for both AbstractSampled or two AbstractCategorical)

[tiemvanderdeure]

The reason it returns this NoLookup currently is because nothing else is implemented for promote_dims/promote_first (I think?), so it would make sense to kind of mirror that.

[rafaqz]

Yeah. Its a bit odd to have the same thing in two places... I wonder if we can somehow make it share the same code.

[rafaqz]

Like have this comparison dependent on the results of promote_first