Feature/managed memory

[fluidnumerics-joe]

This PR brings in the following major changes

• Host and Device memory management is now handled with HIP managed memory. This is done to prepare for MI300A architectures. Testing has found that there is minimal impact to kernel performance
• Core interpolation and differentiation routines are now implemented with HIPBLAS. This improves performance of grid interpolation and differentiation operations significantly. Details will be provided in comments below.
• Variable dimension has been swapped with element dimension. Vector dimension, and matrix rows/cols dimension have been pushed past variable dimensions. This will allow us to take advantage of improvements in feq-parse where operations can now be done on multidimensional arrays for equation evaluation.