fix: increase max_tokens to 2048 and add robust markdown JSON cleaner for NVIDIA SLM

fix: execute explanation generation on CPU without 90s GPU scheduling delay

fix: prioritize text json dataset for demo scenarios in app.py

perf: remove top-level GPU invocation to prevent ZeroGPU startup queueing delay

fix: decorate _generate_explanation with @_gpu_decorator for ZeroGPU runtime

fix: register ZeroGPU compatibility startup hook in app.py

fix: default SLM_MODEL_NAME to openai/gpt-oss-20b for NVIDIA Build API

fix: replace binary dataset with text demo_scenarios.json and enable dynamic RAZOR Q&A

feat: update app.py with user launch configuration and CORS origin logging

fix: restore standard demo.launch loop with strict_cors=False for HF Space persistence

fix: enable FastAPI CORSMiddleware in app.py to resolve cross-origin preflight requests

fix: update merchant state property names in app.py

feat: complete RazorShield AI risk platform