[FEA] Handle `NaN` for floating point types in `MIN`/`MAX`/`ARGMIN`/`ARGMAX` aggregations

[ttnghia]

Currently, the aggregations MIN/MAX/ARGMIN/ARGMAX aggregations use the default/buildit < comparison operator for finding the min/max row in the input column. For floating point types, we may need some other special handling, such as always considering NaN as the maximum value. We should be able to handle such comparisons.

In addition, since not everybody wants that behavior, we can implement such behavior as an option to the aggregation classes and the users can get what they want (either comparing NaN resulting UB or NaN is always the maximum value).