Verify bitcode can pass `llvm-dis` in tests

[artemdinaburg]

Sometimes we generate bitcode that does not pass verifier. As a sanity check, lets ensure that we can always disassemble the bitocode via llvm-dis as a part of our test suite.

[surovic]

I think we don't need to explicitly use llvm-dis for this, the roundtrip tests should be enough, because they call clang. The recent issue we had is more about the fact that our current tests did not trigger it. In hindsight, I should've asked for / added a regression test before merging #89

[artemdinaburg]

I don't mind calling llvm-dis since it would be nice to have the bitcode available for debugging any issues as well (its normally a step i do manually); and it may help narrow down problems between bitcode encoding and triggering other clang failures.