Running a warm bubble experiment end to end in ICON4Py, which is what the
cycle 37 bets are collectively for: a physics driver that carries TMX and
muphys, a land model to close the surface, output that survives more than one
rank, and validation that does not depend on Fortran serialized data.
why is this jot working
Every cycle 37 pitch belongs here except the kernel optimisation work, which is
performance across the dycore and is not about this experiment.