The filename from which a DB first gets opened (either via an explicitly
specified filename, or via the path sequence now configurable at the script
layer) is now "sticky", meaning re-opening won't switch to a different file.
This was easiest by moving most state into the MMDB class itself. The previous
approach of tracking the two DB instances via a smart pointer and blowing the
pointed-to objects away as needed is now instead one of two objects fixed over
the lifetime of Zeek, able to open/close/reopen their underlying Maxmind DBs.
The MMDB class now only has one Lookup() method since there was no need to break
them apart -- it saves the return of a MMDB_lookup_result_s over the stack and
there's no need for throwing an exception.
