Now that we have some LLVM IR generated, we can run PassManager on our
module to get a bunch of neat optimisations on it. Try it out with 3 + 2
to see some constant folding.
Note that the original tutorial uses FunctionPassManager which optimises
on a function per function basis: llvm-hs doesn't expose this yet (and
this is all using the legacy pass manager anyway), so for now we just
optimise the entire module at the end.