openai/gpt-chat-latest → B (3:2)

Side B fixes a semantic inconsistency by making `thread_post_index` consistently 0-based for rank history, replacing silent `unwrap_or(0)` fallbacks with `expect(...)` to enforce an invariant, updating documentation, rendering, and adding integration tests that verify the behavior. Side A improves maintainability by replacing manually enumerated Kaocha test namespaces with a single auto-discovered `^test\..+` suite, but it is primarily a test configuration simplification rather than a functional correctness change.