added a comment - - edited
I think I reproduced this already. Seems to be related to PostgreSQL + IDENTITY only. The SequenceIdentityGenerator must be used but apparently it is not used.
Thanks for the report. Will keep you updated.
ps. The standalone reproduce scripts are good. Thanks for that! If you are interested it would be even better if the test code would be provided as a unit test or with some additional comments and/or assertions about what the desired behavior is or where the supposedly wrong behavior is (even if it is very obvious, like in this issue).