Add cuda-oxide J2K encode stage kernels

[jcwal1516]

Summary

• add a cuda-oxide J2K encode device crate covering deinterleave, RCT/ICT, DWT 5/3, DWT 9/7, and quantize kernels
• add feature/env-gated runtime routing via cuda-oxide-j2k-encode and J2K_CUDA_USE_OXIDE_J2K_ENCODE
• generalize cuda-oxide build staging so nested builds use the staged pinned nightly instead of the outer Cargo toolchain

Validation

• cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-encode
• cargo clippy -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-encode
• CUDA box jcwal@100.75.125.59: J2K_CUDA_OXIDE_ARCH=sm_89 J2K_REQUIRE_CUDA_OXIDE_J2K_ENCODE=1 J2K_CUDA_USE_OXIDE_J2K_ENCODE=1 J2K_REQUIRE_CUDA_RUNTIME=1 J2K_REQUIRE_CUDA_KERNEL_BUILD=1 NVCC=/usr/local/cuda/bin/nvcc cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-encode (95 passed)