[BUG] Incorrect result for COUNT_VALID/COUNT_ALL in rolling_window

[mythrocks]

In rolling_window() operations, COUNT_VALID/COUNT_ALL should only return null rows if the min_periods requirement is not satisfied. For all other cases, the count produced must be valid, even if the input row is null.

As it currently stands, the COUNT* rolling_window() operation returns null if even one of its input rows is null. That behaviour, while correct for aggregations like SUM, is incorrect for COUNT.

E.g. Consider a vector with all nulls:

[null, null, null, null, null]

1. COUNT_ALL with window (preceding=2, following=1, min_periods=1) should yield [2, 3, 3, 3, 2], not [null, null, null, null, null].
2. COUNT_ALL with window (preceding=2, following=1, min_periods=3) should yield [null, 3, 3, 3, null], not [null, null, null, null, null], because min_periods is not met at either end of the vector.
3. COUNT_VALID with window (preceding=2, following=1, min_periods=1) should yield [0, 0, 0, 0, 0], not [null, null, null, null, null].
4. COUNT_VALID with window (preceding=2, following=1, min_periods=3) should yield [null, 0, 0, 0, null], not [null, null, null, null, null], because min_periods is not met at either end of the vector.

The following test should illustrate the expected results, and serve to reproduce the erroneous output:

TEST_F(BorkedCountWindowTest, TestCount)
{
  using T = int32_t;
  using namespace cudf::test;

  auto const input_col = fixed_width_column_wrapper<T>{
    {0, 1, 2, 3, 4, 5}, 
    {1, 0, 1, 1, 1, 1}}.release();

  auto const input_size    = input_col->size();

  auto const preceding   = 2;
  auto const following   = 1;
  auto const min_periods = 3;

  expect_columns_equal(
    cudf::rolling_window(input_col->view(),
                         preceding,
                         following,
                         min_periods,
                         cudf::make_count_aggregation(cudf::null_policy::INCLUDE))->view(),
    fixed_width_column_wrapper<T>{
      {-1, 3, 3, 3, 3, -1},
      { 0, 1, 1, 1, 1,  0}
    });

  expect_columns_equal(
    cudf::rolling_window(input_col->view(),
                         preceding,
                         following,
                         min_periods,
                         cudf::make_count_aggregation(cudf::null_policy::EXCLUDE))->view(),
    fixed_width_column_wrapper<T>{
      {-1, 2, 2, 3, 3, -1},
      { 0, 1, 1, 1, 1,  0}
    });
}

It would be good to have a Pandas perspective on this assessment.