Investigate whether C-code creation and caching works as expected

[erikvansebille]

We heard that in some cases, scipy-parcels is (much) faster than JIT-parcels; especially when very many short calls to pset.execute() are performed. It may be that the caching of the C-kernel is not working as expected, so that every new call to a JIT execute with the same kernels comes at a relatively big overhead. This needs to be checked