Allow `FaceDivFree` with a refinement ratio of 4

[BenWibking]

Currently, FaceDivFree only works with a refinement ratio of 2:

amrex/Src/AmrCore/AMReX_Interpolater.cpp

Line 29 in 4d09691

* The algorithm is restricted to ref ratio of 2.

This limits our MHD simulations to ref_ratio of 2 on all levels.

While not critical, this would improve performance at scale on Frontier, especially when running with MLMG, by allowing us to drop from 8 levels to maybe 5 or 6 levels.

[cgilet]

We would also use this feature at CMP and LSC.

[WeiqunZhang]

It's not ideal. But it seems that should be able to do it in two steps of FillPatch. Create an intermediate "level" using the coarsened-by-2 version of the fine BoxArray.

[cgilet]

I thought @jbbel offered to derive something for this.

[BenWibking]

I thought @jbbel offered to derive something for this.

That was also my recollection from the HPSF Q&A.

[jbbel]

Candace Gilet and I are working on it

[BenWibking]

I wrote this to do it recursively inside the interpolator itself, but may be unneeded now: #5226.