add support for amrex::FFT::PoissonOpenBC for self-gravity

[BenWibking]

Describe the proposal
There is a new FFT-based Poisson solver for open boundary conditions in AMReX. It may provide better performance because it can explicitly compute the Green's function in Fourier space and cache it for re-use each timestep. (The Green's function only has to be computed once.)

The downside is that the Green's function requires as much memory as one component of the level 0 multifab. At extreme scales, FFT can also require more communication, but it needs to be tested to see which one is faster in practice.

Describe alternatives you've considered
Keep using the real-space amrex::OpenBC solver for self-gravity.

Additional context
Documentation: https://amrex-codes.github.io/amrex/docs_html/FFT.html#poisson-solver

If someone works out the math for the Green's function, we could also write a custom solver to do mixed periodic-open boundary conditions for solving the Poisson equation for "tall box" simulations:

FFT::PoissonOpenBC is a 3D only solver that supports open boundaries. Its implementation utilizes FFT::OpenBCSolver, which can be used for implementing convolution based solvers with a user provided Green’s function. If users want to extend the open BC solver to 2D or other types of Green’s function, they could use FFT::PoissonOpenBC as an example.

[BenWibking]

@markkrumholz @lizmcole This project might also be of interest.