Add benchmark results and visualization

Add DGX Spark benchmark results for gpt-oss-20b

Lower GPU memory utilization to 0.90

Switch to NVIDIA vLLM container for Blackwell support

Update DGX Spark specs: GB10 (sm_121), not B200

Remove unnecessary torch dependency

Add platform-specific PyTorch sources for DGX Spark support

Add test matrix table to README

Initial commit: vLLM Blackwell benchmark setup