SNOW-1653121: Support some Timedelta aggregations on axis=0.

[sfc-gh-mvashishtha]

Fixes SNOW-1653121

Test and support aggregation on axis=0. We still raise NotImplementedError if:

1. the aggregation requires concatenating a frame with timedelta types
2. the aggregation requires transposing a row containing a timedelta type and other types.

To facilitate this change, introduce the class SnowparkPandasAggFunc, which represents a Snowpark pandas aggregation, together with the bool preserves_snowpark_pandas_type. We can use SnowparkPandasAggFunc instead of the separate mappings GROUPBY_AGG_PRESERVES_SNOWPARK_PANDAS_TYPE and GROUPBY_AGG_WITH_NONE_SNOWPARK_PANDAS_TYPES.

This change also fixes the bug that timedelta aggregations like mean would produce the wrong type (and the wrong result) by truncating the float result if preserves_snowpark_pandas_type.

[sfc-gh-mvashishtha]

marking as draft till I fix SNOW-1654730

[sfc-gh-nkrishna]

Nice work, LGTM!