Model card: add context window & prefill sizes; base_model = VibeThinker-3B (128K)

Emberglass model card: in-browser WebGPU Qwen2.5-3B + runtime LoRA hot-swap

initial commit