Aggregation coherence with ANSI SQL and query engine integrations

[myrrc]

Vortex has some specified behavior for aggregation functions, but it's not formalized. Duckdb and Datafusion as our main query engines have other specified behaviour. Our vortex-duckdb and vortex-datafusion integrations have a third set of behavior. Goal of this epic is to bring our integrations to the main query engine processing.

NaN means both + and - nan since both Duckdb and Datafusion treat both the same.

Integer (BIGINT / i64)

Query	ANSI	DuckDB	DataFusion	Vortex	Vortex + Duckdb
sum([1,2,NULL])	3	3	3	3	3 (pushed)
min / max([1,2,NULL])	1 / 2	1 / 2	1 / 2	1 / 2	1 / 2 (pushed)
avg([1,2,NULL])	1.5	1.5	1.5	1.5	1.5 (pushed)
count([1,2,NULL])	2	2	2	2	2 (pushed)
sum(all-null)	NULL	NULL	NULL	0	MISMATCH 0 (#9084)
min / max(all-null)	NULL	NULL	NULL	NULL	NULL
avg(all-null)	NULL	NULL	NULL	NaN	MISMATCH NaN (#9083)
count(all-null)	0	0	0	0	0

Floating point

Floating point pushdown is turned off since #9069 but we can revert that for all functions but max.
Therefore last column is the situation where we push to floats

Query	ANSI	DuckDB	DataFusion	Vortex (skip_nans)	Vortex + Duckdb
count([NaN,1,2,NULL])	impl-def	3	3	2	3
sum([NaN,1,2])	impl-def	NaN	NaN	3	NaN
min([NaN,1,2])	impl-def	1	1	1	1 (not pushed)
max([NaN,1,2])	impl-def	NaN	NaN	2	NaN
avg([NaN,1,2])	impl-def	NaN	NaN	1.5	NaN
count([NaN,NaN])	impl-def	2	2	0	2
sum([NaN,NaN])	impl-def	NaN	NaN	0	NaN
min([NaN,NaN])	impl-def	NaN	NaN	NULL	NaN (not pushed)
max([NaN,NaN])	impl-def	NaN	NaN	NULL	NaN
avg([NaN,NaN])	impl-def	NaN	NaN	NaN	NaN
sum([inf,-inf,0])	impl-def	NaN	NaN	NaN	NaN
min / max([inf,-inf,0])	impl-def	-inf / inf	-inf / inf	-inf / inf	-inf (not pushed) / inf
sum([inf,1,2])	impl-def	inf	inf	inf	inf
sum(all-null)	NULL	NULL	NULL	0	MISMATCH 0
min / max(all-null)	NULL	NULL	NULL	NULL	NULL
avg(all-null)	NULL	NULL	NULL	NaN	MISMATCH NaN
count(all-null)	0	0	0	0	0

String (VARCHAR / Utf8)

Query	ANSI	DuckDB	DataFusion	Vortex	Vortex + DuckDB
min / max(['banana','apple',NULL])	apple / banana	apple / banana	apple / banana	apple / banana	apple / banana (pushed)
count(['banana','apple',NULL])	2	2	2	2	2 (pushed)
min / max(all-null)	NULL	NULL	NULL	NULL	NULL
count(all-null)	0	0	0	0	0
sum(x) / avg(x)	type error	error	error	unsupported	not pushed