There is a business requirement not to reject the entire batch.
We experimented with relaxing PAIN schema validation, but found that the code does not support that. (The result is unacceptable: a "Failed" batch will be interrupted at the point of failure, AFTER some transactions have completed successfully. The customer would have to review; just fixing the bad data and resubmitting could end up with duplicate transactions.)
We understand that IBM is working on a solution; in the mean time, this business requirement will not be met.