core: Meditate on the attribute names situation.

[PapyChacal]

As brought up in:

• core: Implement MemrefLayoutAttr base class and use in parser. #2718
  We currently annotate the base Attribute class with a name ClassVar. I.e., it is expected that any subclass defines a name - including further attribute base classes, used as interfaces for example.
  The only practical issue with this I've observed is in IRDL verifiers, that actually uses this transparently. Then, using such a base class errors if it has no name.
  More on the principles side of things, I feel they should all define a name if we so say in the base class indeed. So we could:
• Not expect all of them to have a name anymore, but only at the irdl_attr_definition stage
• Or just enforce indeed that they all have names.

[superlopuh]

My take on this is that we don't want inheritance in IRDL, if we can avoid it, so the approach in #2718, with an override of a class variable, is not ideal. This is separate to our Python implementation, where inheritance seems like the most ergonomic approach. What I'm trying to say is that it would be great to use this task to track a discussion of how to define abstract attributes and use them in constraints, and also how to map these to PyRDL.