This makes sure that the updated delegate exists in the database and that its optimistic locking version hasn't changed.

After this change the ConcurrencyTest with JPA Map storage completes.

Design changes:

• common base class for all JpaDelegateProviders to write the code only once. Pushing up the delegate instance variable. To enable the comparison interface JpaRootEntity introduced
• ClientService will filter out invalid clients as before.
• Interface didn't match well for logoutAll(), therefore therefore catching the exception here - ideas welcome
• ExportUtils will fail if a client can't be exported ... this changes the existing behavior, but seems to be reasonable as exportutils should export everything, and no clients shuold be missing.

Optional stuff (please comment):

• Entity manager could be pushed to JpaDelegateProvider as well.

Will need to do the same changes in #9746 (Client Scope for JPA Map) as well.

Testing this yesterday, I found that the optimistic locking failed when updating only some attributes: the version in the parent wasn't incremented as the the ManyToOne relation is owned by the child.

I looked for a simple solution to change that ownership, but didn't succeed. I then added some hibernate listeners and iterfaces to trigger the essential session.lock(root, LockModeType.OPTIMISTIC_FORCE_INCREMENT).

I added a second commit to this PR to keep them separate for an initial review and comment. After initial feedback on the approach I'll then merge them.

The approach is inspired by https://vladmihalcea.com/jpa-entity-version-property-hibernate/ and https://vladmihalcea.com/hibernate-event-listeners/.

I tested it manually by changing role attribute or changing a client's "Backchannel Logout Session Required"

@vramik, @sguilhen, @hmlnarik - I'd be happy if you could have a look.

Thanks @vramik for the findings, I update the PR with once commit that contains all of your changes. To make it simpler for @sguilhen and @hmlnarik to review, I kept the commits separate. That should make it simpler to decide if both steps should be taken.

After a conversation with @hmlnarik the changes about the hibernate listener described in #9761 (comment) are no longer part of this ticket; they will be covered in a future ticket. Work on that will start once this has been merged.

Previous commit still available here: 4c0bcee21d106a8451b89ec94c122bf75ac0d8e8

@sguilhen - Thanks for the review. I updated this PR according to your review comments. Please have another go if this is good to go. Thanks!