[WIP] Add amrex::LaunchRaw

[AlexanderSinn]

Summary

This PR aims to provide a unified interface to be able to write kernels using shared memory and __syncthreads for CUDA, HIP and SYCL without the need to use ifdefs.

The number of threads per block is always a compile-time known 1D value, while the number of blocks can be 1d, 2d or 3d using the build-in platform indexes like blockIdx.y etc.

TODO:

• Perf testing for some existing kernels
• porting/simplifying existing kernels to use this
• Write documentation
• add tests

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