feat: implement IQP encoding benchmark and reference against Torch

[vvvdwbvvv]

Related Issues

#1227

Changes

• Bug fix
• New feature
• Refactoring
• Documentation
• Test
• CI/CD pipeline
• Other

Why

This change is the first stage of the IQP kernel roadmap toward matching or exceeding the torch.compile reference implementation. We need a reliable torch reference for IQP behavior, stronger hard-case numerical coverage, and a benchmark that makes CUDA-vs-torch parity measurable before deeper kernel optimization work.

How

• added a reusable torch IQP reference implementation for both iqp and iqp-z
• added an IQP benchmark comparing QDP CUDA, eager torch, and torch.compile
• added Python GPU correctness tests with hard single-sample, batch, and IQP-Z cases
• strengthened Rust IQP integration tests around FWT-path and hard-case coverage
• updated benchmark utilities and docs for IQP / IQP-Z support
• adjusted qumat_qdp imports so pure-Python helpers remain importable when _qdp is not built

Checklist

• Added or updated unit tests for all changes
• Added or updated documentation for all changes

[vvvdwbvvv]

This PR is bigger, feel free to suggest.

[ryankert01]

Duplicated with #1189

[ryankert01]

Feel free to reopen it if otherwise

[vvvdwbvvv]

Okay I'll do phase encoding!

[ryankert01]

@vvvdwbvvv Please do it after my PR get merged! Thanks!

[400Ping]

@vvvdwbvvv @ryankert01 's PR has been merge. I will help reopen this

[vvvdwbvvv]

Alr, Got it! Will continue to finish it.

[ryankert01]

draft it to avoid early review. also I suggest to have a fresh start instead.

[rich7420]

hey @vvvdwbvvv — i think your iqp torch reference here will be useful for something i'm planning.

a plan.lower("pennylane") lowering layer in qdp (~2 months out) would diff iqp-encoded states against a known-good torch reference for correctness — and yours in #1231 fits that exactly.

no urgency, no pressure. just a heads-up so we can coordinate signatures when you're ready to leave draft. a ping then would be appreciated.