Slice 6: GPU integration — magnify_motions_gpu(), tests, docs

[joeljose]

Context

Parent: #11 | PRD: #10 | Design: docs/design/gpu-acceleration.md

Goal

Wire all GPU passes together into magnify_motions_gpu(), add comprehensive tests, update docs, and prepare for release.

Acceptance Criteria

• magnify_motions_gpu(data, magnification, width, nlevels, biort, qshift, device) orchestrates Pass 1 → Filter → Pass 2
• main() calls magnify_motions_gpu() when --gpu is set (replaces "not yet implemented" stub from Slice 2)
• GPU path prints GPU name, VRAM, batch sizes at startup
• End-to-end: python motion_mag.py --gpu -i face.mp4 produces valid output
• tests/test_motion_mag_gpu.py with all tiers:
  • Tier 1: memory estimation
  • Tier 2: forward/inverse roundtrip, phase extraction shapes, batched consistency, filter shapes
  • Tier 3: full pipeline smoke test on face.mp4
• All existing CPU tests still pass
• README updated with GPU usage section (--gpu, Docker GPU setup, VRAM requirements)
• CHANGELOG updated
• VERSION bumped to 2.0.0

Blocked by

#16 (Slice 5) — needs all GPU passes implemented.

Files likely involved

• motion_mag.py — magnify_motions_gpu() orchestration, main() integration
• tests/test_motion_mag_gpu.py — new file
• README.md — GPU section
• CHANGELOG.md — v2.0.0 entry
• VERSION — bump to 2.0.0

Testing approach

• ./test.sh — all CPU tests pass
• ./test.sh gpu — all GPU tests pass
• Manual: compare GPU output video against CPU output for visual quality
• Manual: test on different video resolutions

[joeljose]

Completed in PR #18, released in v2.0.0