Some sources of data, e.g. ASA NSEL, record a bidirectional flow record
with separate counters for in/out bytes and in/out packets. This patch
makes aggregation over "packets", "bytes", "bps" etc look at the *sum* of
the input and output counters. Previously they would use only the input
counters.
Query types which explicitly look at one direction, like "ibyte" and
"obyte", are unchanged.
Processing of unidirectional flows is not affected, because these store
values in the "in" counters only - the "out" counters are zero, so using the
sum of in+out is the same.
The resulting code is simpler: it doesn't have to calculate both in/out
values everywhere. It also uses accessor functions for all values, which
means no special cases for r->counter[cindex].
