Retain the pre-frame size check and allocation budget while avoiding untracked input-sized buffers. Add the reported heap regression and container/progressive compatibility checks. Co-authored-by: Codex <noreply@openai.com>
