Relational persistence provider does not cleanup data when a user transaction is rolledback off a different thread

When storing data in relational databases and using user transactions, if a user transaction is rolled back from a different thread than the thread which created the transaction and which was used by ModeShape to detect the transaction, the backend database connection and transient data is not cleared correctly.

This can be reproduced in a specific test case which uses Arjuna and forces the reaper thread to abort timed-out transactions. See linked forum issue for more details