Some improvements and testing for the EB covered state

[baperry2]

This closes #703, with the goal of ensuring that covered cells never affect fluid cell values:

• A change in behavior for all simulations - if a NaN is detected in the state after a timestep, the simulation now stops
• For all EB stencils, if a value in the stencil is zero, that element of the stencil is ignored rather than multiplying it by the value in the cell and adding it into the gradient being computed. This should not effect the solution, but is done so that if the value is NaN in a covered cell, it does not propagate into the domain.
• A new function is added to abort if the stencil for gradients at the boundary tries to access covered cells (can happen for the quadratic option on coarse grids ebd.boundary_grad_stencil_type = 0)
• the function zero_in_body has been deleted because it was not used anywhere
• A new option to set the body state to 0 for all variables for debugging (to ensure this does not impact fluid cells)
• The eb-c3 test is modified to set the body state to 0. This is done to prevent any future changes that would cause data from covered cells to be accessed during the fluid cell updates. Had to make a few changes to Tests/CMakeLists.txt because this test can't run with AMReX FPE trapping.

[hsitaram]

looks good! thank you, Bruce.

[baperry2]

The end of timestep NaN check works at least on AMD and Nvidia GPUs (Frontier and Eagle).

On GPU, the abort for invalid stencils only works in debug mode or with USE_ASSERTION=TRUE.

Did not notice a change in performance.

[jrood-nrel]

There is a typo here.

[jrood-nrel]

This test won't have these options amr.plot_file=plt amr.checkpoint_files_output=0 amr.plot_files_output=1 amrex.the_arena_is_managed=0, which I think we would want to enforce.

[jrood-nrel]

Actually I'm confused. It just has the same options as the other tests?

[jrood-nrel]

You could leave the RUNTIME_OPTIONS as is and just add amrex.signal_handling=0 to this test category.