Improve AMReX diagnostics and kernels

[marchdf]

This PR:

• add diagnostics to AMReX implementation (diagonals and grid summary)
• aligns pressure init with other implementations
• improves kernel: use local temporaries for repeated array values. The compiler has trouble seeing through the repeated array values and triggers more global memory reads than necessary. This leads to a 1.15X speedup.

This is one of several PRs stemming from a discussion with @sjsprecious and @johnmauff.