Update benchmarks

Updates benchmark cases: NonUniformGridCPU, NonUniformGridGPU, UniformGridCPU, UniformGridGPU, PhaseFieldAllenCahn, and Thermocapillary.

The updates concern a general clean up, and extension of the benchmarks to support:

• MachineState: for increased reproducibility of benchmarks
• Mixed-precision: This means the LBM benchmarks can be generated using float32 for storing the PDFs and float64 for the collision step, BC, etc
• The NonUniform benchmarks are aligned now with the benchmarks of Schornbaum. This means the weak and strong scaling benchmarks are done using a specific LDC setup

Furthermore, the mesh refinement is extended in the case an inplace streaming pattern is used. In this case, the sweepCollection provides a blockwiese streamCollide and ghostLayerPropagation, and the communication kernels provide a blockwise equal level local communication. This enables better performance when smaller blocks are used.