Instead of storing both a list and a map, we now only store maps of
entity/subdomain fragments -> unique identifiers in order to reduce the
memory overhead. Moreover, the public interface of the entity networks
is changed such that one can no longer get fragments of a sub-domain
with a specific id as the overall map already contains this information.
This can be added in the future as an external function to extract data
of a specific subdomain from the network. The builder classes are also
restructured to no longer store unneccesary data. Only the public
interfaces of entity networks have changed, and due to the small
user-base, and as that interface is currently only used internally by
the writer (which has been modified), we don't guarantee backwards
compatibility here.