Summary

• switch managed Codex config model from gpt-5.3-codex to gpt-5.4
• add long-context defaults in managed config: model_context_window = 1050000 and model_auto_compact_token_limit = 945000
• set plan_mode_reasoning_effort = "xhigh" to match default development reasoning effort
• keep entrypoint fallback config aligned with auth-sync managed config
• add tests for codex config creation/rewrite invariants

Verification

• pnpm --filter ./packages/lib test tests/usecases/auth-sync.test.ts
• pnpm --filter ./packages/lib typecheck
• pnpm --filter ./packages/lib test

Closes #109
Closes #111