This was a stupid decision. We will have to assign the
network at some other place. The problem will be a problem
when we have two gsm_networks in one process and the same
subscriber is traveling in both networks.
gsm_subscriber is now refcounted, the db backend is leaking
a lot less, db_get_subscriber will allocate the subscr record
now, subscr_* will look up a subscriber in the list of currently
active subscribers and add an ref to this one.
The db test cases pass, more testing will be when next to the bts