[codex] Add /context window breakdown

[starr-openai]

Summary

• add a new /context TUI command that requests a semantic context-window breakdown from the app-server
• expose that breakdown through a new thread/context/read app-server API and protocol schema
• render the returned context map in the TUI, including loading state and verbose support
• fix thread-local replay so async /context results return to the originating thread instead of the currently active one
• add focused coverage for the thread replay path and developer-content classification path

Why

Codex already tracks token usage, but it did not provide a way to inspect what was actually occupying the current model-visible context. This change adds a user-facing inspection path and closes a correctness gap in the async TUI result handling.

Validation

• cargo test -p codex-core developer_message_content_items_are_classified_individually
• cargo test -p codex-tui context_window_breakdown_for_inactive_thread_replays_into_origin_thread
• just fix -p codex-core
• just fix -p codex-tui
• just fmt