refactor: modernize type hinting by using collection ABCs and built-in generic types across the codebase

fix: convert all run_pipeline runtime errors to gr.Error consistently

fix: surface mask-validation ValueError as gr.Error in run_pipeline

fix: cap working fps and surface prewarm failures (critique of b859009)

fix: list-audit fixes — LaMa duration, multi-stream audio, prewarm wait

refactor: round 7 — collapse dual mask-empty check into _get_mask_from_editor

refactor: round 5 — share crop/mask computation, tighten tempfile, faster PNG

refactor: round 4 — relocate composite_and_save, top-level pipeline imports

refactor: round 3 — stdout-pipe first-frame extract, _card_title, _msg helpers

refactor: round 2 — add CropRegion edge/box properties, externalise CSS

refactor: round 1 — drop dead code, lift constants, simplify dispatch

feat: stream VACE chunks, prewarm checkpoint cache, per-chunk progress

feat: snap-to-rectangle button + wire LaMa to LAMA_MODEL_URL mirror

feat: implement VACE-14B quality mode

feat: auto-trim long uploads and add audio re-encode fallback

feat: add mask-clear button and reject too-small / too-large watermarks

.

feat: introduce checkpoint mirroring script, strengthen video validation, and improve pipeline robustness for masking and compositing.

refactor: implement streaming LaMa inpainting, optimize crop aspect ratio matching, and add support for high-frame-rate video presets

Initial pipeline upload