feat: in-app example moments + refined prompt on the Space

revert(space): back to Qwen2.5-VL-7B after 3B failed verification

feat(space): switch to Qwen2.5-VL-3B (small-model ethos, Tiny Titan)

fix(backend): first pass respects refined env flag; add A/B override

feat(space): wire Qwen2.5-VL-7B via hf_local provider (stage 2)

build(space): adopt proven Gradio launch pattern for ZeroGPU

build(space): stage 1 — deployable shell for HF Space

feat: avoid repeating films across a session

feat(backend): resilience — timeouts, bounded retry, clean errors

feat(backend): sanitize and re-encode all input before the model

feat(server): serve React UI and interpret API from gr.Server