As for a work around, only way to get around it would be to go to an earlier version of Paratext, like 9.5.101.3.
We introduced a new library that is stricter on malformed Unicode, and this revealed an existing bug where Paratext was splitting surrogate pairs when doing word analysis.