Clarify the Struct::field API

[gatesn]

Currently, all encodings that support the struct DType implement the StructVariant trait. This primarily allows accessing the fields of the struct array by index and by name.

However, the null-semantics for this are not well defined when both the struct array and the field have validity masks.

Perhaps it should be StructVariant::maybe_null_field, or perhaps the Struct::field should return an array with the intersection of the validity.

The latter option would mean that calling StructArray::field(0) on {a: i32}? could return i32? and not necessarily the field DType i32 that might be expected.