Avoiding overflow in Simulation.num_cells

[momchil-flex]

For wrong settings, it was possible to get overflow in the Simulation.num_cells even with 64-bit precision integers.

[yaugenst-flex]

I guess this relates to the min grid spacing validator. from #2335

[momchil-flex]

Well, kind of, but do those validators only check absolute bounds? In the problematic simulation I was investigating, the grid spacing was as if for ~1um wavelength, but the simulation domain was on the order of tens of mm in each direction. Neither of these is prohibited on its own, but the combination produces an immense amount of grid points. If it were not for overflow making the total number of cells negative, the simulation would have hit various validators, at the very least the pre-upload one.