[iamnbutler]

Adds Retry to a failed run (map ^525, ^526; decided in Q154).

• When the chat's latest run failed and nothing has run since, a line under the composer reads "The last run failed." with Retry, for anyone who may invoke the agent. Retry asks the agent Try again. with the chat's current pick.
• When the error is the model's or its key's (401/403/404, API key, unauthorized, not found, no access…), the line says so ahead of time: "The last run failed on openai/gpt-4. Retry switches this chat to Automatic." Retry then sends the ask with a new automatic flag, which makes the channel replace the chat's model with its own choice (the host default, or the first usable model), resets the composer's pick to Automatic, and confirms in a toast.
• ask gains automatic?: boolean in @ace/channel/protocol; older hosts ignore it and keep the chat's model.

Smoke-tested on a loopback host (port 4432, isolated home/config/keychain) with headless Playwright: ace ask retry --model openai/gpt-4 failed with OpenAI's 404; the app showed the Automatic line, Retry showed the "Switched this chat to Automatic" toast, the line went away, and the chat answered Hello! on claude-opus-5-5.

Built in Ace

No. Built with Claude Code as a fix-train worker against the Ace map, outside Ace (#5).