new method for temporary FABs

[zingale]

Originally we used Elixirs for temporary FABs.

Then we switched to doing a resize, like:

FArrayBox  tmp;

...

     // inside the MFIter loop
    tmp.resize(bx, 1, The_Async_Arena());

Now AMReX can do something like:

auto tmp_fab = amrex::FArrayBox{bx, ncomps, amrex::The_Async_Arena()};

This is the new stream-ordered allocator.

We can should look into updating the code with time.

[zingale]

see https://amrex-codes.github.io/amrex/docs_html/GPU.html#async-arena

There was some suggestion at HPSF that this could be more efficient.

[WeiqunZhang]

The new approach is cleaner. But the old approach also works and it uses the stream-ordered allocator as well. One advantage of the old approach is it is better for OpenMP performance.