Minor: Add tests showing aggregate behavior for NaNs

[alamb]

Which issue does this PR close?

Related to #10627 and #8031

Rationale for this change

While looking at #10627 from @westonpace I couldn't find any existing tests for how Nans worked in aggregates.

What changes are included in this PR?

Add a test that documents how Nans work today in the basic aggregates

Are these changes tested?

Only tests

Are there any user-facing changes?

No

[alamb]

it is surprising to me that the aggregates on f32 seem to handle Nan consistently with postgres, but the aggregates on f64 do not (I would expect max(col_f64_nan), avg(col_f64_nan), sum(col_f64_nan) to all return NaN

[comphead]

What concerns me is below:

DataFusion CLI v38.0.0
> CREATE TABLE float_table (
     col_f32 FLOAT,
     col_f32_nan FLOAT,
     col_f64 DOUBLE,
     col_f64_nan DOUBLE
 ) as VALUES
 ( -128.2,  -128.2,                      -128.2,  -128.2 ),
 ( 32768.3, arrow_cast('NAN','Float32'), 32768.3, 32768.3 ),
 ( 27.3,    27.3,                        27.3,    arrow_cast('NAN','Float64') );
0 row(s) fetched. 
Elapsed 0.028 seconds.

> select min(col_f32), max(col_f32), avg(col_f32), sum(col_f32), count(col_f32) from float_table;
+--------------------------+--------------------------+--------------------------+--------------------------+----------------------------+
| MIN(float_table.col_f32) | MAX(float_table.col_f32) | AVG(float_table.col_f32) | SUM(float_table.col_f32) | COUNT(float_table.col_f32) |
+--------------------------+--------------------------+--------------------------+--------------------------+----------------------------+
| -128.2                   | 32768.3                  | 10889.13359451294        | 32667.40078353882        | 3                          |
+--------------------------+--------------------------+--------------------------+--------------------------+----------------------------+
1 row(s) fetched. 
Elapsed 0.011 seconds.

but in both duckdb and postgres

>
>  select -128.2 col_f32,  -128.2 col_f32_nan,                      -128.2 col_f64,  -128.2 col_f64_nan union all
>  select 32768.3, 'NaN'::DOUBLE PRECISION, 32768.3, 32768.3 union all
> select 27.3,    27.3,                        27.3,    'NaN'::DOUBLE PRECISION) x
> ;
┌──────────────┬──────────────┬────────────────────┬───────────────┬────────────────┐
│ min(col_f32) │ max(col_f32) │    avg(col_f32)    │ sum(col_f32)  │ count(col_f32) │
│ decimal(6,1) │ decimal(6,1) │       double       │ decimal(38,1) │     int64      │
├──────────────┼──────────────┼────────────────────┼───────────────┼────────────────┤
│       -128.2 │      32768.3 │ 10889.133333333333 │       32667.4 │              3 │
└──────────────┴──────────────┴────────────────────┴───────────────┴────────────────┘

[comphead]

AVG is incorrect precision

[alamb]

AVG is incorrect precision

One reason the AVG result may different is that your example used single precision (f32). When I run the same query with double precision (f64) DataFusion gets the same 10889.133333333333 as duckdb

> select min(col_f64), max(col_f64), avg(col_f64), sum(col_f64), count(col_f64) from float_table;
+--------------------------+--------------------------+--------------------------+--------------------------+----------------------------+
| MIN(float_table.col_f64) | MAX(float_table.col_f64) | AVG(float_table.col_f64) | SUM(float_table.col_f64) | COUNT(float_table.col_f64) |
+--------------------------+--------------------------+--------------------------+--------------------------+----------------------------+
| -128.2                   | 32768.3                  | 10889.133333333333       | 32667.4                  | 3                          |
+--------------------------+--------------------------+--------------------------+--------------------------+----------------------------+
1 row(s) fetched.
Elapsed 0.001 seconds.

[comphead]

lgtm thanks @alamb
however I think there is some issue with UNION ALL and type derivation

> select avg(a) from (select -128.2::float a union all select 32768.3 union all select 27.3);
+--------------------+
| AVG(a)             |
+--------------------+
| 10889.133334350587 |
+--------------------+
1 row(s) fetched. 
Elapsed 0.011 seconds.

> select avg(a) from (select -128.2::float a union all select 32768.3::float union all select 27.3::float);
+-------------------+
| AVG(a)            |
+-------------------+
| 10889.13359451294 |
+-------------------+
1 row(s) fetched. 
Elapsed 0.011 seconds.

I expect those numbers to be the same. And that what duck db does

[comphead]

Filed #10688