Codex should switch to Read-only Planning Mode when quota runs out

Unfortunately for those meta tasks codex is still going to be using inference and reasoning so a quota of tokens and interactions is still going to be needed. But it’s an interesting idea because it may use less in those phases (or it may not!)