Add comprehensive AI test dataset and CI/CD improvements

[MatrixA]

Summary

Add 25 AI-generated test fixtures covering 12 tools across image/video/audio formats, with 48 new integration tests and restructured CI/CD pipeline. Total test count grows from 62 to 115.

Changes

• Add 14 image fixtures: GPT Image (2), Midjourney (4), Ideogram (1), Flux (2), SeedReam (4), Nano Pro (1)
• Add 9 video fixtures via Git LFS: Sora (2), Kling (3), Vidu (1), SeedAnce (1), LTX (1), Wan (1)
• Add 2 audio fixtures: Suno (1), ElevenLabs (1)
• Create tests/image_detection.rs with 26 tests (24 active + 2 deep/ignored)
• Create tests/video_detection.rs with 20 tests (18 active + 2 deep/ignored)
• Create tests/audio_detection.rs with 6 tests
• Configure Git LFS via .gitattributes for tests/fixtures/*.mp4
• Restructure CI into 3 jobs: lint (fast, no LFS), test (with LFS + ffmpeg), test-deep (ignored deep-mode tests)
• Add LFS fixture validation step to catch pointer-only checkouts
• Install ffmpeg in CI for video frame watermark detection

Checklist

• cargo fmt -- --check passes
• cargo clippy -- -D warnings passes
• cargo test passes
• New detection methods include appropriate confidence tiers
• Documentation updated (if applicable)