Add image-only pipeline mode

Expose encoder input resolution as score_size (default 224)

Per-stage eval budgets image_max_evals/video_max_evals (200/100)

Split into image_num_generations (50) / video_num_generations (25)

Replace num_candidates/max_evals with num_generations (population_size + num_generations)

Replace num_rounds with population_size (default 20), matching clean_search

Align two-stage search with canonical: clean scoring, art_data stage-1, image/video batching

Two-stage pipeline (image search then motion search) with per-stage budgets

ROI-enhanced search space + docs; EBA caption

Publish self-contained NEVO stimulus-synthesis pipeline + model card