experiment with AMReX WorkEstType

[zingale]

Previously we had an option to do a separate distribution map in the reactions, but that was removed in #877.

AMReX added the WorkEstType() in 17.10:

 -- Added a load balance approach to Amr/AmrLevel.  The runtime
     parameters for this is amr.loadbalance_with_workestimates and
     amr.loadbalance_level0_int.  The latter is for single level
     (i.e., amr.max_level=0) only.  For multi-level runs, load balance
     is done during regrid and thus the load balance interval is
     controlled by amr.regrid_int.  To use this approach, AmrLevel
     derivative must has a StateData for work estimates and its index
     is returned by virtual function int WorkEstType().

We can try using this to balance based on the number of RHS evaluations. However, I think the reason it was removed originally is because on a GPU this may not be much of a win, since all of the threads will wait for the last zone to burn.