Add scalability section + reframe limitations

Add comprehensive model card with eval results

Upload 200-step GRPO LoRA adapter (peak 0.506, +283% vs base)

feat: GRPO-trained Qwen2.5-3B for clinical trial auditing (50 steps, T4)

initial commit