EvolveScalarWave3D on GPU

[nikwit]

As a first simple minimal example, I managed to get EvolveScalarWave3D to run on the GPU without copying any data back and forth except for I/O.

Restrictions:

• Only works for aligned meshes (i.e. Brick)
• Only works on a single CPU core as pupping device pointers and sending them to a different charm process is not safe without a deep copy (should be solved by nodegroup runtime system hopefully)
• global time stepping with fixed time step only
• no outer boundaries (periodic boundary conditions)