Add cuda-oxide HTJ2K compact kernel

[jcwal1516]

Summary

• add a cuda-oxide j2k_htj2k_compact_codeblocks entrypoint to the J2K encode PTX artifact
• route only HTJ2K encoded-byte compaction through J2K_CUDA_USE_OXIDE_J2K_ENCODE; entropy encode kernels stay on CUDA C
• add a strict CUDA fixture for assembled HT cleanup payload compaction and update env/unsafe docs

Validation

Local macOS:

• cargo fmt --check
• cargo fmt --check --manifest-path crates/j2k-cuda-runtime/src/cuda_oxide_j2k_encode/simt/Cargo.toml
• git diff --check
• cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-encode (94 tests; cuda-oxide PTX skipped on macOS)
• cargo clippy -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-encode -- -D warnings
• cargo test -p xtask --test repo_lint
• cargo xtask unsafe-audit

CUDA host jcwal@100.75.125.59 / RTX 4070 SUPER / sm_89:

• cargo oxide build --arch sm_89 in crates/j2k-cuda-runtime/src/cuda_oxide_j2k_encode
• strict cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-j2k-encode with 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 (96 tests)
• combined strict all-oxide cargo test -p j2k-cuda-runtime --all-targets --features cuda-oxide-copy-u8,cuda-oxide-j2k-encode,cuda-oxide-j2k-decode-store,cuda-oxide-j2k-dequantize,cuda-oxide-j2k-idwt,cuda-oxide-transcode (105 tests)
• cargo +nightly clippy --manifest-path crates/j2k-cuda-runtime/src/cuda_oxide_j2k_encode/simt/Cargo.toml -- -D warnings