Another attempt to fix weird non-deterministic initialization issue

Another attempt to fix a weird issue where Euler integration inexplicably and non-deterministically fails (large integration error) due to an unusually large flux at the boundary after the first time step. The most logical explanation still seems to be that uninitialized array variables (which can be small floating point numbers, infinite, or zero) are screwing up the initial step size. a1b68fcf attempted to fix this but it doesn't seem to have worked, so now we also ensure that diagnostic variable caches are initialized to zero. I haven't observed the issue again after a few runs, so we'll see if this indeed fixes it.