docs: define bidirectionality design intent - #6047

Problem

Astryx has RTL implementation guidance and browser audits, but no current human-owned record defining what correct bidirectional behavior means. Reviewers can see whether pixels changed without having one authority for what should mirror, what should remain physical, and how interaction meaning should change.

Changes

Add a draft design:bidirectionality record covering:

• semantic mirroring rather than mechanical reflection;
• content direction versus layout direction;
• exactly-once treatment of directional artwork;
• physical and direction-neutral exceptions;
• keyboard, drag, scroll, overlay, and responsive behavior;
• mixed-direction text and browser dir semantics;
• pointer/touch targets staying aligned with visible controls and preserving their owning component's target-size requirement.

The touch-target requirement incorporates the review lesson from #5177: proving center alignment alone is not enough. Target size remains owned by AST-020 and the applicable component/composition contracts.

This stays separate from audit mechanics and the component-audit rubric. The design record says what people should experience; evidence architecture says what proves it; component contracts adopt the relevant requirements.

This PR is intentionally draft and remains non-authoritative until design-owner approval. Its architecture references must be promoted before, or together with, this record.

Test plan

• node scripts/check-knowledge.mjs
• git diff --check
• Verified promotion requirements against the current knowledge schema.
• Cross-checked the touch-target requirement against fix(core): center coarse-pointer touch targets in RTL for Checkbox, Radio, and Switch (#5170) #5177 and AST-020.