Implement framework-agnostic extraction logic for component interface

Summary

Add a shared extraction utility that partitions a serialized tag's children into three categories: properties (meta tags with property attribute), named refs (top-level children with data-name attribute), and anonymous content (everything else). This utility is framework-agnostic and will be consumed by each framework's renderer.

Acceptance Criteria

• New extractComponentInterface(tag) function in packages/transform/src/helpers.ts (or new file)
• Returns { properties: Record<string, string>, refs: Record<string, SerializedTag[]>, children: RenderableTreeNode[] }
• Properties extracted from children where attributes.property is set, keyed by property name, value from attributes.content
• Refs extracted from top-level children where attributes['data-name'] is set, keyed by data-name value
• Nested refs (children inside other refs) are NOT extracted — they stay inside their parent ref
• Remaining children (neither property nor named ref) returned as anonymous content
• Unit tests covering: basic extraction, nested refs stay nested, empty cases, mixed children

Approach

1. Add extraction function to packages/transform
2. Iterate children once, partitioning into three buckets based on attributes
3. Export for use by framework renderers

Resolution

Completed: 2026-04-04

Branch: claude/adr-008-implementation-nBN9K

What was done

• Added extractComponentInterface() and fromKebabCase() to packages/transform/src/helpers.ts
• Added ComponentInterface type export
• Exported new functions from packages/transform/src/index.ts
• 11 unit tests in packages/transform/test/extract-interface.test.ts

Notes

• Uses data-field attribute on meta tags (not property) since that's what createComponentRenderable actually produces
• Returns camelCase keys from kebab-case data-field values via fromKebabCase