Add CUDA 13 support for DGX Spark

- Add Dockerfile.dgx_spark_cuda13 with NGC PyTorch 25.11 (CUDA 13.0, PyTorch 2.10)
- Rename Dockerfile.dgx_spark to Dockerfile.dgx_spark_cuda12
- Fix GraphBolt CMakeLists.txt for CUDA 13 (sm_120/sm_121 support)
- Update build_dgx_spark.sh to support both CUDA versions via parameter
- Update documentation (README.md, docker/README.md, docs/DGX_SPARK.md)

Tested: DGL Core, Sparse, GraphBolt, GNN layers all working on CUDA 13

docs: Add verified build status for DGX Spark with sparse and graphbolt

- Build tested with BUILD_SPARSE=ON and BUILD_GRAPHBOLT=ON
- Verified DGL Core, Sparse, and GraphBolt modules working on GB10
- Build time: ~6 minutes with CUDA 12.8, PyTorch 2.6, sm_120