enable MQM for empty strings / missing content

Extract out of Arles email:

Just a quick note for something to consider in future development. I noted a number of cases recently where annotators marked up single letters at the start of segments. The cases were things like “Señor Vasquez…” and the reason was that they wanted to tell us that it should be “El Señor Vasquez…” but annotations must contain at least one character. For a number of our error types, however, there is no logical reason for them to have content (e.g., Omission). Maybe in the future we could change things to allow empty spans? I don’t see this as a high-priority issue at the moment, but it is something to consider.