docs: add GPT 5.3 Codex benchmark results (-45.1% token savings)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>