Kernel optimization

[huiyuxie]

The kernel optimization has already marked as TODO in this repo.

Check the below tasks:

• Remove control flow divergence from boundary_flux_kernel!
• Refactor set_diagonal_to_zero! in cuda_volume_integral! and check the precision using IdealGlmMhdEquations
• Change set_log_type("log_Base") and set_sqrt_type("sqrt_Base") to other ways
• See whether a prototype for create_cache is needed or not
• Possible change from isequal to simple math expression and compare performance
• Use synchronization to combine close kernels
• Combine two similar kernels in mortar kernel

[huiyuxie]

Possible change from isequal to simple math expression and compare performance - this one need further check with benchmark especially for 3D problem.