tfm_spm_free_service_handle() doesn't clear the content of the service node (mainly the rhandle).
This introduced a bug when running PSA crypto tests which allocates/free services many times, a new test which didn't set rhandle yet re-uses a service node from a previous test in which the node-rhandle data is not 0, so get_rhandle returns a non-NULL pointer which is actually a dangling pointer.