Add ZeroGPU backend so any HF model runs on Space GPU (default XHToken/Spark-X2.5-4B)

Fix unsupported-model retries and error message: fail fast when no Inference Provider serves the model

Fix Space pipeline failures: g++ via packages.txt, chat_completion with fallback, clear auth errors, Windows exe fix, non-gated default model

Remove unsupported placeholder from Dropdown

Add searchable model dropdown

Initial commit: Algorithm Finder Gradio app