tolerances for green line global experiment verificaiton

[halungge]

For the verification of the global model some tolerances (mostly rtols ) had to be released in order for the solve_nonhydro timestep to verify. Not all of them are parallel to tolerance bounds in the blueline code and should be revised and verified whether it is ok or still due to a undiscovered bug.

There are #TODOs in the code (test_solve_nonhydro.py) to track these issues (see #315)

Originally posted by @halungge in #315 (comment)