Original comment by @LuddeWessen:
I was about to post an issue report on this as well, since I need to put constraints on choice variables, especially between them.
I believe this should be possible for at least two reasons:

If the parameters have internal ordering we often need to put constraints as if it were range parameter.
Sometimes categorical or ordered parameters needs to be constrained to remove "symmetries" in the parameter space, e.g. when parameters (1,2) actually means the same as (2,1), we want to disallow one of them.
I believe that the fix suggested by @Balandat Jul 17 sounds good.

Proposed solution: #621 (comment)