fix: sync batch size with sample size in gradio app

fix: Increase timeout

feat: Add batch size configuration in api and frontends

feat: Add registry to support any kind of inference provider

docs: Add clarifying text to gradio app

perf: cache blbl schema at module load instead of rebuilding per request

cleanup: remove spaces import and @spaces.GPU decorators (hardware changed to cpu-basic)

fix: increase @spaces.GPU timeout to 300s to avoid GPU task abort on slow LLM calls

fix: catch exceptions in do_evaluate to show error in UI instead of crashing ZeroGPU runtime

fix: add @spaces.GPU decorator to satisfy ZeroGPU spaces check; graceful fallback when spaces not installed

fix: disable SSR mode in Gradio launch to prevent Node.js server crash on HF Spaces

feat: Add Gradio app for HF Spaces deployment