Visual regression testing for GenerationPanel

[thedavidweng]

Problem

GenerationPanel is the most complex UI component. Refactoring it (issue #53) risks visual regressions with no automated detection.

Scope

• Set up Playwright screenshot comparison for the GenerationPanel
• Capture baseline screenshots for key states (idle, generating, completed, error)
• CI gate: pixel diff threshold < 0.1%

Acceptance criteria

• Playwright captures screenshots of GenerationPanel in key states
• CI fails on visual regressions exceeding the threshold
• Baseline screenshots are easy to update when changes are intentional