ENH: Add support for min_count keyword for Resample and Groupby functions

[phofl]

• closes BUG: downsampling with last doesn't accept "min_count" keyword #37768
• closes BUG: Groupby.last accepts min_count but implementation raises #37821
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

This would add support for the documented keyword ```min_count``. If we decide to move forward with this, we have to decide what to do in the following case:

df = DataFrame({"a": [1] * 3, "b": [np.nan] * 3})
result = df.groupby("a").func(min_count=0)

first and last return 0.0 which is fine I think and consistent with sum, but min returns inf while max returns -inf.

Default parameter is set to 1 to avoid backwards incompatible changes. We should keep this at least until 2.0

If we decide to not implement min_count here, we should adjust the docs and the function signatures for the groupby functions.

cc @rhshadrach

[rhshadrach]

Thanks for the PR @phofl. It seems to me that first, last, min, and max should all return nan even when min_count is 0. Sum behaves differently because the empty sum is 0 (likewise, the empty product is 1).

[rhshadrach]

Default parameter is set to 1 to avoid backwards incompatible changes. We should keep this at least until 2.0

Do you perhaps mean -1? Otherwise, I don't follow.

[DiSchi123]

My 2 cents: min_count=0 is the same as not specifying a min_count. Min_count of x says you need to have at least x non-nan values in the bin to show a result, otherwise show nan. If you have a min_count of 0 isnt that like saying show me the sum, first etc regardless how many nans are there? Maybe I am missing something?

But while it is important to decide what min_count=0 should show it is less important for the implementation, you would not specify that imo, because as mentioned it is like not specifying it. I am ok with any decision.

[phofl]

cc @rhshadrach When we want to return nan in this case, we can keep -1 as default.

[rhshadrach]

Overall looks really good, just some minor comments on tests.

[rhshadrach]

cc @jreback @jbrockmendel

[jbrockmendel]

can you avoid the double lookup/comparison here by setting min_count = max(min_count, 1) outside the loop?

[phofl]

Done

[jreback]

lgtm. ping on green (resolved conflict)

[phofl]

@jreback greenish. Failure is unrelated

[jreback]

thanks @phofl