Distinguish nils in firstBuffer

[angelamayxie]

Fixes dolthub/dolt#9035

In AggregationBuffer.firstBuffer , the buffer should not update if the first row has already been written. However, there was no way of distinguishing between a nil because the buffer is empty and a nil because the value in that column of the first row is nil.

When you have two rows (NULL, 1) and (1, NULL), the first row gets written into []firstBuffer.val as {nil, 1}. Because there's no way to tell whether the first value is supposed to be nil, it ends up getting overwritten by the 1 in the second row. Because the second value is not nil, Update skips it. As a result, we end up with a row {1, 1} that matches neither of the original rows and evaluates to TRUE when grouped by c0 = c1, even though the original rows evaluate to NULL.

This isn't an issue when the rows are (NULL, 1), (NULL, 1) or (1, NULL), (1, NULL) because the nil value remains a nil value when Update is called on the second row.

I fixed this by adding a writtenNil flag to firstBuffer to indicate that the nil value is meant to be there and should not be overwritten.

[jycor]

Nice catch!
LGTM