Rework importing when using Locations to be not as slow.

Some changes are made to AbstractLocation and the URL model to support this. First, the identity_hash attribute and set_identity_hash() method are moved from URL to AbstractLocation to support bulk operations across location types. AbstractLocation now calls set_identity_hash() as part of its clean() routine. Finally, a bulk_get_or_create() classmethod is added to AbstractLocation; it is responsible for (hopefully efficiently) bulk getting-or-creating an iterable of the appropriate type (e.g., URLs). It works by querying for existing locations based on identity_hash values, and only creating new ones using Django's bulk_create() method.

Both LocationManager and EndpointManager have been changed. Of greatest impact, LocationManager has been updated to use the same accumulator pattern that EndpointManager was recently updated to use (thanks Val!). While doing this, I noticed the opportunity to consolidate things a bit; rather than maintaining the V3 if/else structure throughout, I standardized around the set of methods used by EndpointManager. At the same time, I inserted another level of indirection -- LocationHandler, which under the covers delegates to one or the other manager -- to help keep the two managers more in line going forward. The intent is, we won't be able to introduce one-off features for only Endpoints or Locations without at least considering how to handle the other. Plus, it removes a lot of the "#TODO: DELETE ME IN THE FUTURE MAYBE" comments right now. Additionally, LocationProductReference status is now kept in sync with finding-level status changes during reimport (if any finding ref is Active, the product ref is Active; otherwise Mitigated). This was a pre-existing gap.

The general process was, take the excellent pattern set by EndpointManager, and beat it until it worked with Locations too.

In my very unscientific testing of a 3k findings import, V2 finished in ~21s, while V3 finished in ~23s, putting Locations solidly within tickling distance of Endpoints.