Don't create redundant dimensions in group-specific terms

[tomicapretto]

If we have (1 | group) and (predictor | group) and then we decide to use aliases such as alias_1 and alias_2 the inference data will contain alias_1__factor_dim and alias_2__factor_dim. But they actually refer to the same levels, which are the groups in group.

Would it be possible to remove this redundancy? What should be the name of the dimension?