Introduce return_data_type for Aggregate function

[jayzhan211]

Is your feature request related to a problem or challenge?

return_data_type is not the same as the data_type for List case. For example, data_type: List(Int64), return_data_type will be Int64

Ref: https://github.com/apache/arrow-datafusion/pull/7242/files

Describe the solution you'd like

Introduce return_data_type for Sum, and other Aggregate function

Describe alternatives you've considered

No response

Additional context

No response

[alamb]

I think we should do some more careful thinking of how to handle multi-phase aggregates where the intermediate type is not the same as the output type -- for example SUM(x) for non lists can use the same aggregate (as you can sum partial sums) but COUNT(x) does not -- the first phase must actually COUNT but the second needs to use SUM.

This causes problems for #6937 where we might want to dynamically stop doing the first phase aggregation.

[jayzhan211]

TODO

• min_max (single-phase)
• median (two-phase, 1. collect values 2. compute median)
• count (no need, data type is always i64)
• avg (had been done)

[jayzhan211]

Reopen it if there is any concrete issue to solve