[Webtoon study] Improve clean-image prompts for lettering and transition-aware panels

[realproject7]

Context

The samples show that clean images must leave room for lettering and often need to serve as reaction or transition panels rather than complete standalone illustrations. Current image prompts emphasize scene art, but not enough about balloon-safe composition, silence, object inserts, or vertical scroll timing.

Required Changes

Update clean-image prompt generation for cartoon cuts:

• Add letteringSafeZones to cut planning.
• Require prompt text to reserve quiet areas for speech/thought/narration overlays when needed.
• Support image purposes:
  • establishing shot
  • action impact
  • reaction close-up
  • object insert
  • UI insert background
  • silence/pause panel
  • transition image
  • cliffhanger image
• Prompt for panel aspect and whitespace intent, not just shot type.
• Allow agents to request noImageTransition instead of generating unnecessary art.

Data Model

Add optional cut fields:

• imagePurpose
• letteringSafeZones[]
• cameraDistance
• transitionRole
• silenceWeight
• expectedOverlayDensity
• negativeTextPolicy

Acceptance Criteria

• Generated prompts include clear balloon-safe composition instructions when overlays are expected.
• Reaction and object-insert cuts are supported as first-class prompt modes.
• The cut review can display the intended image purpose.
• Existing image generation/import flows still work.

Verification

• Prompt-builder tests for each image purpose.
• A generated original episode includes at least one object insert, one reaction close-up, and one silence panel.
• Exported lettering does not routinely cover faces or key objects in the test story.