refactor: restructure CLAUDE.md with progressive disclosure pattern

[twjeffery]

Restructured the AI assistant documentation following best practices from
https://www.hlyr.dev/blog/writing-a-good-claude-md to improve effectiveness
and maintainability.

Key changes:

• Reduced CLAUDE.md from 375 lines to 87 lines (77% reduction)
• Kept only universally applicable architectural knowledge in CLAUDE.md
• Created agent_docs/ directory for task-specific detailed documentation
• Organized content into 6 focused documents:
  • component_workflows.md - Step-by-step component development
  • pr_testing_guide.md - PR playground usage and best practices
  • testing.md - Testing commands, frameworks, and strategies
  • naming_conventions.md - Comprehensive naming standards
  • common_issues.md - Troubleshooting guide
  • folder_structure.md - Complete directory reference

Benefits:

• Better signal-to-noise ratio for AI assistant
• Progressive disclosure allows reading detailed docs when needed
• Reduced instruction count improves overall instruction-following
• More maintainable and searchable documentation
• All original content preserved, just reorganized

The new structure ensures Claude sees critical architectural patterns
(wrapper pattern, Svelte as source of truth, design tokens first) in
every conversation while keeping task-specific details available on-demand.