Calculate network stream and filtered bandwidth averages per relay type

After documenting better Torflow in #40056 (closed), is now more clear that Torflow is calculating these averages per relay type (ie. guard, exit, guard+exit and middle) (see https://gitlab.torproject.org/tpo/network-health/sbws/-/blob/cde77e9dd9c9b35b403b34e0e8deda63438857a3/docs/source/torflow_aggr.rst, Stream and filtered bandwidth for all relays and https://gitlab.torproject.org/tpo/network-health/sbws/-/blob/cde77e9dd9c9b35b403b34e0e8deda63438857a3/docs/source/differences.rst, Network stream and filtered bandwidth), while sbws is not.

This will influence the scaled values (#40051 (closed))