Different behavior of distinct count on null inputs

[viirya]

Describe the bug

A few Spark tests with distinct count failed while working on #250: https://github.com/apache/arrow-datafusion-comet/actions/runs/8681807652/job/23805034669?pr=250

[info] TwoLevelAggregateHashMapSuite:
[info] - multiple column distinct count *** FAILED *** (530 milliseconds)
[info]   Results do not match for query:
...
[info]   == Results ==
[info]   !== Correct Answer - 1 ==   == Spark Answer - 1 ==
[info]   !struct<>                   struct<count(key1, key2, key3):bigint>
[info]   ![3]                        [4] (QueryTest.scala:243)

Spark distinct count aggregation doesn't count null inputs. I.e.,

override lazy val updateExpressions = {
  ..
  Seq(
    /* count = */ If(nullableChildren.map(IsNull).reduce(Or), count, count + 1L)
  )
}

But seems DataFusion count aggregation function behaves differently.

Steps to reproduce

No response

Expected behavior

No response

Additional context

No response

[viirya]

I'm working on this to find out the root cause.