We currently maintain a manually written segmenter for emoji segmentation to determine emoji presentation style: text or emoji.

In order to improve correctness, maintainability and security, we should move this to a grammar based implementation without manually maintaining the scanner state machine.