Never suggest agents fetch sign-in codes from a mailbox

CLI output told agents they 'may fetch the email code' if they had inbox access with permission. In a test, Claude searched the only mail connector in the session, which belonged to a different person, before noticing and stopping. Agents do not reliably check that a connector belongs to the user first. Agent-facing text should say to ask the person for the code and not to look in a mailbox.