Require a node's cache to take "ownership" of an SFSB context before returning it.

A cache that creates the SFSB context can take ownership without any cluster-wide message.
A cache that has taken ownership can lock it during subsequent requests. Doing this does not require any cluster-wide message.
A cache that doesn't have ownership of an SFSB context must request ownership via a cluster-wide message. This will be granted, but not until any concurrently executing requests on the owning node have completed.
If the node owning the session does not respond to the request to give up ownership (e.g. it has crashed) ownership is acquired.

Code for this locking function exists in AS; needs to be extracted out into ha-server-api (JBCLUSTER-223)

The same concept will be applied to web sessions.