I'll share what I know with you in the hopes that it will help somehow.
Well KC (keycloak) is a super-set of the PL (PicketLink) functionality thus in theory it ought to work fine once it is ready and once some sort of migration path is known. You may not wish to move to KC due to the additional functionality which may be off-putting for lite applications but KC will perform everything PL did and more and will do so in VM memory if you so choose.
Essentially KC is a real federated authentication and authorization service with identity management that can run standalone or in-VM within a WildFly cluster. Although a Java implementation it works with other systems and languages out of process. It does integrate with Spring which may interest you.
The following link provides information for Wildfly 9 clustered installation:
Thus you should be able to have your authorization demands met _in VM_ as opposed to over-the-wire for performance reasons if necessary.
IMOP I think the KC project is the right move. They are fixing the big issue which is the lack of an opensource Federated Identity Management System. They also fixed little things such as Composite Roles which are missing from PL.
I merely disliked the abrupt change-over. I also can't move to keycloak until I have more of an idea how the migration would work.
For example, how different is the default KC relational schema from the default basic PL schema:
It is also not clear if keycloak has a CDI demand system ready like PicketLink. They only hint at it. Also it runs in-cluster on Wildfly 9 and I am on 8. Nothing huge but issues that will need to be addressed.
Hope that helps.