[MAJOR FEATURE] Add CUDA version of nonrigid HO array-based optimizer.

[Algomorph]

Phase 1: design / sandbox / CMake

• add CMake to make CUDA targets if nescessary
• determine (near) optimal data structure
• determine optimal transition points between RAM & VRAM

Phase 2: implement basics

• implement initial data structure
• template a few math function and see how it works
• figure out how to offload data for testing

Phase 3: design alterations / detailed implementation & tests
(More detailed & complete list coming soon)