Improve caching of ORCA jobs

[danielhollas]

After a brief testing of caching, it seems to work! 🎉

With one caveat: One needs to use the same number of CPUs per ORCA CalcJob. It would be really great to lift this restriction, but it might be complicated since orca encodes the nCPUs in its input file. Probably should discuss with aiida-core team first, and then possibly fix this in the aiida-orca plugin.

As a short-term improvement, we should be properly loading up number of MPI tasks when loading the workflow from the database.

[danielhollas]

Looks like number of MPI processes in ORCA is passed manually via a PAL section of the input dictionary. So I guess we need to have a separate non_db input node so that it can be excluded from hash calculation. Needs to be supported in aiida-orca first.