[Roadmap] Strengthen CUDA kernel implementation

[ryankert01]

Our current implementation of compute kernel is slower than torch.compile() implementation. This Roadmap aims to at least match with torch.comile implementation.

Stage 1. Add torch implementation reference

• #1189 @ryankert01
• phase @vvvdwbvvv

Stage 2. Strengthen our CUDA kernel

• amplitude encoding @ryankert01
• angle encoding @400Ping
• iqp encoding @vvvdwbvvv
• phase @vvvdwbvvv
• etc.

These PRs should

1. strengthen current numerical correctness test with hard test cases
2. strengthen CUDA kernel implementation accordingly
3. make sure numerical correctness test with hard test cases passes
4. make sure we are faster or match torch.compiles speed

[vvvdwbvvv]

I can give you some help!