fix: move @spaces.GPU to app.py for ZeroGPU detection

fix: top-level @spaces.GPU for ZeroGPU detection

fix: ZeroGPU @spaces.GPU support for model generation

feat: ZeroGPU (RTX Pro 6000) support with 8B model

fix: robust thinking tag handling for MiniCPM5-1B

fix: robust thinking tag handling for MiniCPM5-1B

fix: remove token_type_ids for MiniCPM5-1B compatibility

fix: handle MiniCPM5-1B thinking tags, extract content from <think>

fix: model generation reliability + performance

fix: remove device_map='cpu' - not supported in HF Spaces

fix: use transformers+MiniCPM5-1B, fix session_id State capture

fix: use ctransformers (pre-built wheel) instead of llama-cpp-python

feat: working pipeline with MiniCPM5-1B GGUF on CPU

feat: initialize DreamLand project scaffold