CLI crashes with “server_error: The server had an error while processing your request” when context window is nearly exhausted

When using the @openai/codex CLI interactively with a very large conversation history, once the API context window is nearly full (e.g. “1% context left”), the CLI throws an internal server error and exits rather than gracefully warning or trimming the context.

## Reproduction Steps

1. Install the latest CLI:

``bash
npm install -g @openai/codex
``

1. Start an interactive session:

``bash
codex
``

1. Enter or paste a very long prompt (or build up many turns) until you see:

``
— 1% context left
``

1. Continue chatting or issue another prompt.
2. Observe the CLI crash with a stack trace similar to:

``text
file:///…/codex/dist/cli.js:443
…pg.NEWLINE_REGEXP=/\r\n|[\n\r]/g;function Fwe(e,t){…}
…
[Error]: The server had an error while processing your request. Sorry about that!
at e.a [as iterator] (…/cli.js:443:1514)
at process.processTicksAndRejections (node:internal/…)
at async I0.run (…/cli.js:460:2356) {
status: undefined,
headers: undefined,
request_id: undefined,
error: {
type: 'server_error',
message: 'The server had an error while processing your request. Sorry about that!',
}
}
``

## Expected Behavior

• Detect that the context window is exhausted (or about to be) and either:

1. Prompt the user to trim or clear history (e.g. suggest /clear),
2. Automatically drop the oldest tokens before continuing, or
3. Smart auto-compact mechanism
4. At minimum, emit a clear warning instead of an unhandled server error.

## Actual Behavior
Once the API rejects due to context overflow, the CLI surfaces a raw server error stack trace and exits.

## Environment

• macOS 13.5
• Node.js v20.18.3 (via nvm)
• No custom configuration