Extraction follows recognized emoji forms, not code-point guessing
RGI sequences and their recognized non-fully-qualified forms are kept intact. Text-presentation requests and non-RGI regional-indicator pairs are not silently promoted to standard emoji.