[estebanzimanyi]

A doxygen block whose wrapper backs several SQL functions or aggregates lists them on one tag
line, @sqlfn mergeFinal(), tCountFinal(), tMinFinal() and @sqlaggfn merge(), tCount(), tMin(),
and the operators they carry the same way, @sqlop @p ->, @p ->> beside @sqlfn a(), b(). The
wrapper maps read every name of such a line: _mdb_to_agg gives the wrapper each aggregate it
names, and _mdb_to_sql gives it each SQL function with the operator of the same position.
_SQLFN, _SQLAGGFN and _SQLOP still find the tag; _tag_list reads its values.

Witness: Temporal_tagg_finalfn of MobilityDB carries @sqlaggfn merge(), tAnd(), tCount(), tMax(), tMin(), tOr(), tSum(), tdensity(), tnpoints(), wCount(), wMax(), wMin(), wSum(), and the catalog
stated sqlAgg = ["merge"] for temporal_tagg_finalfn, a binding reading it finding the member in
one aggregate of thirteen.

Measured: over MobilityDB branch sql/aggregate-combines the catalog differs from the one of master
(bb78d0a94) in one field of 14 functions, sqlAgg, each naming every aggregate its tag lists
(temporal_tagg_finalfn 13, taggstate_serialize and taggstate_deserialize 16,
temporal_tcount_combinefn 3, spanset_union_finalfn 2, ...); the aggregates section and every
sqlfn, which keeps the first name, are identical. Two tests in tests/test_sqlaggfn.py read a
listing @sqlaggfn/@sqlfn and a listing @sqlop; both fail on master's parser and pass here. The
suite runs 560 tests with the same 3 failures as master over the same catalog
(test_family and two of test_struct_layout, failing identically on master).

Why: a member shared by several aggregates states all of them, and a generator building an
aggregate from its roles reads sqlAgg to find them; the first name alone made every shared member
belong to one aggregate. The operator of the second listed name reaches the catalog by its position,
as the tag writes it.