video-interaction-mapper: Enforce readable screenshots

[ilin-figma]

Summary

• Add a hard screenshot-readability policy to Video Interaction Mapper.
• Preserve at least 1440 px width for landscape recordings and 900 px for portrait recordings, unless the source is smaller.
• Refuse to generate a storyboard when assets are missing, undersized, over budget, or have not passed the readability gate.
• Remove the degraded embedded-image fallback when upload_assets is unavailable.
• Add regression coverage for minimum width, native-resolution preservation, and impossible upload budgets.

Why

The asset compressor could repeatedly reduce image dimensions to satisfy the upload-size budget, eventually producing screenshots as narrow as 480 px. The workflow also allowed a direct-embedding fallback that encouraged further downscaling. This made UI text in generated walkthroughs tiny and pixelated.

The workflow now lowers JPEG quality only to a defined floor and then fails clearly instead of sacrificing screenshot dimensions.

Impact

Generated interaction walkthroughs retain readable UI detail. If the source or upload constraints cannot produce a readable asset, the skill reports the blocker before modifying Figma rather than silently creating a degraded storyboard.

Validation

• python3 -m unittest discover -s workflow-skills/video-interaction-mapper/tests -v
• Python syntax compilation for all mapper scripts and tests
• Real-video integration check: 22 before/after assets generated at 1440×815
• Confirmed a forced 560 px landscape run is rejected