hessian_eps test hangs randomly in parallel

[Simon Funke]

The dolfin-adjoint test hessian_eps fails randomly when executed in parallel at the very end of the program.

I suspect that the reason is that PETSc objects are not destroyed in a consistent order when running in parallel.

Because I dont have a fix now, I changed the test script to run the test in serial for now (commit b04a91b09f1531c693c9d52b833465f384de3f6e)