Currently we infer the nullability of a DType::Extension as the nullability of the extension's storage DType.

Take as an example the extension dtype, complex64({i: f32, j: f32}). The problem is a nullable complex64 would have to be a different type from a non-nullable complex64, i.e. complex64({i: f32, j: f32}?).

The proposal is to lift nullability of extensions into the DType enum, and force a non-nullable storage dtype. This non-nullable constraint could be lifted later... but for now I think it makes sense.