[codex] Handle stale terminal input after PTY exit

[gangcaiyoule]

Fix stale terminal input crash

What changed

• guard terminal io writes so stale websocket input sent to an exited PTY no longer crashes the whole Hive runtime
• send a terminal error back to the client and close only that broken socket
• add a regression test covering stale io input after PTY exit

Why

A stale terminal websocket could still send input after the backing PTY had already exited. That bubbled up as PTY is not active for run: ... and terminated the runtime process.

Impact

• Hive stays up instead of crashing when a stale terminal tab sends input to a dead run
• the affected terminal connection gets an error and closes cleanly

Validation

• pnpm check
• pnpm build
• attempted a focused websocket regression test locally; the direct runtime fix was validated, but that targeted Vitest case remained timing-sensitive on this Windows environment