docs: Add plan.md and agents.md architecture blueprints to experimental

[Copilot]

Establishes the foundational architecture documentation for evolving Morsely from a single-page practice tool into a gamified, trail-based Morse learning platform.

docs/plan.md

• Full architecture blueprint: Frontend PWA / API Layer / Learning Engine / Persistence with explicit boundary rules
• Domain model: Trail → Section → Lesson → Practice hierarchy with TypeScript interface definitions
• 7-stage practice evaluation pipeline (normalize → evaluate → score → XP → unlock gates → streak → persist)
• XP formula with per-type base points, streak multiplier, perfect-lesson bonus, and quadratic level thresholds
• Versioned REST API contracts (/api/v1/) with DTO shapes
• DB schema proposal: SQLite (dev) → Postgres (prod) via Drizzle ORM, migration-first
• 8-milestone vertical-slice backlog with acceptance criteria and CI gates per milestone

docs/agents.md

• Canonical domain vocabulary table — enforces section/lesson/practice over legacy session/progression/exercise
• Module boundary rules with explicit forbidden-dependency matrix (engine must never import Express/React/ORM)
• Definition of Done checklist gating every merged increment
• Portability constraints: engine exposed via injectable repository interfaces and plain DTOs, making it replaceable with a Java/Spring Boot backend without frontend changes
• Quick-reference protocol for adding new PracticeType variants end-to-end