arrow

GH-36053:  [C++] summarizing a variable results in NA at random, while there is no NA in the subset of data (#36368)

### Rationale for this change

When merging two aggregate states we were failing to use the correct `no_nulls` field.  This field tells us whether we should return `null` if `skip_nulls=False` (if `no_nulls` is false then we return null).

Since we were reading the wrong field we would sometimes emit null even when a column didn't actually have any nulls.

### What changes are included in this PR?

Fixed the bug.

### Are these changes tested?

Yes, I added a new unit test that reproduced this failure quite reliably.

### Are there any user-facing changes?

No.
* Closes: #36053

Authored-by: Weston Pace <weston.pace@gmail.com>
Signed-off-by: Benjamin Kietzman <bengilgit@gmail.com>