Hotfix/stackframe

[maddyscientist]

This PR primarily serves to remove the induced stack frame in a number of kernels

• Use flavor/chirality swizzle trick in non-degenerate Twisted-mass preconditioned clover application kernel
• Laplace and Staggered quark smearing kernel with NVSHMEM (lack of __force_inline__)
• Use SharedMemoryCache to act as virtual registers (Symanzik improved Wilson-flow and STOUT kernels)
  • In doing so we add support for multiple concurrent dynamic cache objects through use of an offset to the base shared-memory pointer
• Reordering of clover derivative kernel to reduce register pressure

Some other minor changes:

• Fix for HIP compilation which was broken by Feature/large kernel args #1364
• Remove some unused legacy Matrix functions

[kostrzewa]

Doesn't seem to break anything on 8xP100 in the tmLQCD HMC.

[weinbe2]

Looks great, thanks @maddyscientist ! Pending Jenkins completing this is good to merge.