switch Strang from T to e integration

[zingale]

This issue serves to track Microphysics AMReX-Astro/Microphysics#496 as it applies to MAESTROeX.

For Strang integration, we will switch from integrating (T, X_k) to integrating (e, X_k) in the reaction networks. This will require us to call the EOS every righthand side call, but testing has shown that this is not a performance issue (and in some cases, we speed up). Further, once this is in, we will template the EOS call to use eos_re_t, which further cuts down on the EOS expense.

Overall this is more accurate than what we were doing. It also reduces the size of the integration system (and linear algebra).

For MAESTROeX however, we might consider adding the option to have the Strang integration variables be (h, X_k), instead of (e, X_k), since enthalpy is more natural for low Mach flows.

The plan is to merge this before 21.03.