account for (xexp, mosek) failure in test_constant_atoms.py

[rileyjmurray]

In a recent PR of mine, I noticed that testing the composite atom xexp(pos(x)) failed with MOSEK in test_constant_atoms.py. I tried to add this (atom, solver) pair to a list of known exclusions, via the KNOWN_SOLVER_ERRORS list already in test_constant_atoms.py. I now see that the way I did this was insufficient, since it didn't work with composite atoms. I'm preparing a PR that will change test_constant_atoms.py to be more robust to this kind of thing.