Add leaf? distinction to TypeKind

[gmac]

What

In this PR, I'd propose formalizing a proper leaf? distinction on TypeKind.

Why?

At present, TypeKind provides a fields? distinction (Object + Interface) that is a bit awkward because it slices across both Composite and Leaf distinctions...

• fields? != Composite, because it omits Union.
• !fields? != Leaf, because it includes Union, which is not a Leaf.

This ambiguity leads to awkward situations:

• Assert matching response shapes in fields_will_merge #4347 should really be checking .leaf? rather than !fields?.
• fields_have_appropriate_selections does not require selections on union types #4351 stems from ambiguity with the fields? distinction (spec validation should be checking composite?).
• As a library author, I currently need my own helper to correctly assess the Leaf distinction.

[rmosolgo]

Thanks for the improvement! I agree this makes sense.