Fix: expand mask to 3 channels before Laplacian pyramid to prevent shape broadcast error

Feature: Laplacian pyramid edge blending + progress bars for image swap

Fix flickering: detect faces every frame, no per-frame enhancement in video

Fix video glitches: OpenCV-only enhance for video, even dimensions for XVID, softer unsharp

Enhance: CodeFormer ONNX + ESPCN SR + u2net_human_seg + CRF18 unsharp video

Perf: ONNX multi-thread, 320px det_size for video, fast-mode frame skip UI

Perf: source face detected once, target faces cached every 5 frames, 720p video processing

Quality: increase output resolution to 2048px, Lanczos upscale, adaptive unsharp mask

Fix: remove MediaPipe (API broken on Py3.13), add OpenCV face enhancement

Fix: use public mirror for inswapper model (gated HF repo fallback)

Fix build: remove basicsr/gfpgan (Python 3.13 incompatible), use onnxruntime CPU

Initial commit: Face & Body Swapper