fix: explain Codex config load errors caused by a deleted working directory

[EugeneTheDev]

Problem

When the agent's working directory is deleted while it runs, Codex fails every config load with a generic message (IJAI-2923, IJAI-3685):

• config/read → -32603 failed to resolve feature override precedence: No such file or directory (os error 2)
• thread/start → -32600 failed to load configuration: No such file or directory (os error 2)

Codex loads project config from its process working directory, which it inherits from codex-acp, so the error says nothing about the real cause.

Change

New src/CodexErrorExplanations.ts: a single list of rules that every app-server request error goes through in CodexAppServerClient.sendRequest. A rule only rewrites the message; the error class, code and data stay, so CodexThreadErrors and app-server recovery work as before.

• Deleted working directory: the agent records its working directory at start and checks whether the process still points at the directory at that path (a directory recreated at the same path counts as deleted). The message becomes:
  Codex cannot read its configuration: its working directory <path> was deleted or replaced after the agent started. Restart the agent.
• Config files hint: moved here from CodexAcpServer.handleError, so it now applies to every request that fails to load config, not only session new/fork. "feature override precedence" is reworded to "failed to load configuration". Errors that require logout are skipped. handleError keeps only the logout action, via requiresLogout() from the module.

Wire shape note

Session new/fork config errors previously reached the client as data: "<message>\n\nCheck…" (a string, because the text was passed as the data argument of RequestError.internalError). They now take the SDK's default path like every other internal error: data: {details: "<message>\n\nCheck…"}. The logout error keeps its string data.