Add cuda-oxide J2K decode store kernels

[jcwal1516]

Summary

• add opt-in cuda-oxide PTX for J2K decode store and inverse-MCT kernels
• route decode store/MCT launchers through cuda-oxide only when the feature and env switch are both enabled
• document the new env controls and unsafe device-kernel audit row

Validation

• cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-decode-store
• cargo clippy -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-decode-store -- -D warnings
• cargo xtask unsafe-audit
• cargo test -p xtask --test repo_lint
• remote CUDA host: cargo oxide build --arch sm_89 in cuda_oxide_j2k_decode_store
• remote CUDA host: strict runtime test with J2K_CUDA_USE_OXIDE_J2K_DECODE_STORE=1 and J2K_REQUIRE_CUDA_OXIDE_J2K_DECODE_STORE=1, 95 passed
• remote CUDA host: combined cuda-oxide feature/runtime test for CopyU8 + J2K encode + J2K decode store, 98 passed