`atan2` precision loss with integer arguments

[neilconway]

Describe the bug

When atan2 is invoked with integer arguments, the input is coerced to Float32 (not Float64 as the comments in the implementation suggest). This results in losing precision unnecessarily and produces inconsistent results with DuckDB and Postgres.

To Reproduce

> select  atan2(1, 1000000);
+--------------------------------+
| atan2(Int64(1),Int64(1000000)) |
+--------------------------------+
| 0.000001                       |
+--------------------------------+
1 row(s) fetched.
Elapsed 0.084 seconds.

> select  atan2(1.0, 1000000.0);
+------------------------------------+
| atan2(Float64(1),Float64(1000000)) |
+------------------------------------+
| 9.999999999996666e-7               |
+------------------------------------+
1 row(s) fetched.
Elapsed 0.006 seconds.

Expected behavior

DuckDB and Postgres both return 9.999999999996666e-7, matching the DataFusion behavior for Float64 input.

Additional context

No response

[neilconway]

take