fix: load LoRA adapter weights to CPU for ZeroGPU startup compat

Switch inference to transformers + bnb-4bit + LoRA for ZeroGPU

Switch from Zero GPU to T4: remove Dockerfile, simplify theme

Day 1 expanded schema: 5 tables, 14B fine-tuned model

fix: Qwen2.5 chat template for fine-tuned model

feat: swap to fine-tuned Q4_K_M model (4.68 GB)

feat: fine-tuned model + GGUF export pipeline fixes

fix: remove nonexistent fine-tuned model reference to prevent double-linking on HF Space

fix: restore GPU offload on Zero GPU (emulation mode handles module-level CUDA)

fix: CPU-only on Zero GPU (model persists in RAM between queries), 8 threads, restore eager load

fix: preload ALL CUDA .so files, add nvidia-cublas-cu12 + nvidia-cusparse-cu12

fix: preload libcudart.so.12 via ctypes (LD_LIBRARY_PATH has no effect at runtime)

fix: resolve CUDA runtime from pip package, add GPU→CPU fallback

Add CUDA library path resolution for Zero GPU Spaces

Switch to CUDA wheel + n_gpu_layers=-1 for Zero GPU (T4)

Optimize for free HF Space: n_threads=2, remove threading timeout overhead

Initial commit: Kasualdad LFED — Phases 0-8 complete