add: Hosted generation for the Space, and retire the ZeroGPU attempt

Revert "revert: Unpin the Space's torch -- the CUDA 12 theory was wrong"

revert: Unpin the Space's torch -- the CUDA 12 theory was wrong

fix: Keep CUDA calls off the dedicated inference thread -- again

fix: Run inference inside @spaces.GPU, where the weights actually exist

fix: Accept the Space's existing CONTROLAI_BACKEND spelling, and stop guessing

add: A CUDA path for the demo Space, behind the existing engine contract

refactor: Collapse four inference backends into one MLX path

fix: Don't route CUDA/ZeroGPU calls through a manually-created thread

fix: Enforce parameter provenance, fix rendering/citations, add tools, harden agent loop

fix: Don't pass the model-holding agent as a @spaces.GPU argument

fix: Raise ZeroGPU call duration budget; drop stale llama.cpp comment

fix: Launch the ZeroGPU probe Gradio app on its own port, not mounted

fix: Move the ZeroGPU probe function to module scope

fix: Register a real @spaces.GPU Gradio event handler for ZeroGPU startup check

fix: Include the actual Gradio/ZeroGPU code changes missed in the last commit

fix: Serialize inference requests to prevent concurrent llama.cpp crashes

revert: Don't block app startup on a warm-up generation

perf: Warm up the model's KV cache during startup, not on the first user request

perf: Preload model on container startup and optimize PyTorch CPU multi-threading

feat: Initial open-source release of ControlAI