fixed refcount leakage when unboxing from cache

[amirmiron]

There is some leakage of refcounts caused in case of netref realization from cache:

a = remote_object.remote_attr
oid = remote_object.remote_attr.____oid__
oid = remote_object.remote_attr.____oid__
oid = remote_object.remote_attr.____oid__
remote_object.____conn__()._remote_root.exposed_getconn()._local_objects._dict[oid][1]

Would show blowed up refcount.
On unboxing of a reply, if an oid already exists in cache, the same netref is given instead of a new one, causing that we would only one single netref object, that upon its destruction we'd only decref the remote by one, instead of by correct amount. fixed by adding an attribute of refcount on the netref and decref-ing accordingly in a loop.
Another option is to remove the caching altogether but I was unsure as to its importance.
One could also add another parameter for Connection's handle_del and to RefCountingColl.decref to decref not one by one. It could be added as default arguments, or as a different method, or change all other calls to be with a number. Not sure how it might affect others, so didn't do any of that(yet?).

[coldfix]

Hi,

good catch!

Can you make it work by sending the count as additional argument for _handle_del instead of sending multiple requests.