Add a ComposerReasoningToggle UI and persist optimistic composerReasoningEffort on thread runtime. Wire setThreadReasoningEffort action to update runtime state and send session config to the server for OpenAI/codex-cli. Extend workspace-default apply flow to include reasoningEffort, add reasoning metadata to provider/connection catalogs and JSON-RPC schemas, and surface provider-model reasoning defaults. Update Codex app-server developer instructions and dynamic tool descriptions. Include new/updated tests for the toggle, model selection, and JSON-RPC behavior.
