Combine back-to-back saxpy operations, as well as saxpy next to xpay

[mbkuhn]

Summary

This is intended to reduce the number of kernels created on GPUs. The calls to these new kernels are only in MLCGSolver because that is what we have been looking at, but they likely could be added to other places in the code. This came out of the Open Hackathon two weeks ago. @marchdf can provide more detail about the performance improvement this was shown to provide.

Additional background

Checklist

The proposed changes:

• fix a bug or incorrect behavior in AMReX
• add new capabilities to AMReX
• changes answers in the test suite to more than roundoff level
• are likely to significantly affect the results of downstream AMReX users
• include documentation in the code and/or rst files, if appropriate
• provide speedup with no expected diffs