Enhance duplicate screenshot detection (hash/perceptual)

[arividar]

Summary

--skip-duplicates currently compares raw PNG bytes for exact equality. This misses near-identical frames and is fragile to encoding differences.

Proposal

Introduce a more robust duplicate detection strategy:

• Compute a fast hash (e.g., xxHash64/SHA-256) of the PNG buffer and compare hashes instead of storing entire buffers.
• Optionally add perceptual hashing (aHash/pHash/dHash on a downscaled grayscale image) to detect visually identical frames within a small threshold.
• Keep exact mode as default; add a flag like --duplicate-method exact|hash|phash and perhaps a --phash-threshold for tolerance (optional step).

Tasks

• Refactor ScreenshotWriter to store last hash instead of last buffer.
• Implement exact hash mode (hash of PNG buffer) as a first step.
• (Optional) Implement perceptual hash mode with a small threshold.
• Add tests for identical vs slightly different frames (unit tests can simulate buffers; perceptual tests can use small sample images).
• Update README/help text to document the behavior.

Acceptance criteria

• With --skip-duplicates, identical frames do not create files; different frames do.
• Tests cover duplicate detection.
• Documentation updated.