readme

Add Expert tier tasks to evaluate multi-step UUID stateful tracking and OpenAPI discovery

Implement strict judge feedback: ambiguous task descriptions, server-side payload clues, increased exploration steps

Enhance episode state isolation and realistic grader hints for max scoring

Initial submission: API Debug Environment