[r] Add randomized test to detect dimnames errors

[bnprks]

I thought I had found another dimnames bug recently, but it turns out it was already fixed in the current main branch. In the mean time, I had already written some randomized tests that will hopefully provide some extra test coverage against future dimnames regressions.

It is a little slow to run, so I've limited it to a number of randomized tests where there's an 80% detection rate on the old bug, which should take approximately 0.1 seconds.