Add cuda-oxide J2K IDWT kernels

[jcwal1516]

Summary

• add opt-in cuda-oxide J2K generic IDWT PTX build and runtime dispatch
• port generic inverse-DWT entrypoints to cuda-oxide: interleave, horizontal/vertical 5/3 and 9/7, generic batched interleave+horizontal, generic batched vertical, and the legacy single-entry kernel
• leave cooperative shared-memory IDWT kernels on built-in CUDA C PTX for a separate focused PR
• document J2K_CUDA_USE_OXIDE_J2K_IDWT and J2K_REQUIRE_CUDA_OXIDE_J2K_IDWT

Validation

• local: cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-idwt
• local: cargo clippy -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-idwt -- -D warnings
• local: cargo test -p xtask --test repo_lint
• local: cargo xtask unsafe-audit
• local: cargo fmt --check
• local: cargo fmt --check --manifest-path crates/j2k-cuda-runtime/src/cuda_oxide_j2k_idwt/simt/Cargo.toml
• local: git diff --check
• CUDA host: standalone cargo oxide build --arch sm_89
• CUDA host: strict runtime test with J2K_REQUIRE_CUDA_OXIDE_J2K_IDWT=1, 95 passed
• CUDA host: combined strict runtime test with copy-u8, J2K encode, decode-store, dequantize, and IDWT cuda-oxide features required, 100 passed