array_has function returns null for an empty list ([]) instead of false

[bert-beyondloops]

Describe the bug

When using the array_has function with an empty list, the result is

Currently :

array_has([], 1) => null.
array_has(null, 1) => null.

In my opinion, it should only return null when the list itself is null.

In duckDb:
array_has([], 1) => false.
array_has(null, 1) => null.

To Reproduce

CREATE TABLE foo ( bar INTEGER[] );

INSERT INTO foo VALUES 
( [1, 3, 5]),
(  []),
( null)
;

SELECT bar, IFNULL(CAST(list_contains(bar, 1) AS VARCHAR), 'null') FROM foo;

will result in :

+-----------+-----------------------------------------------+
| bar       | nvl(array_has(foo.bar,Int64(1)),Utf8("null")) |
+-----------+-----------------------------------------------+
| [1, 3, 5] | true                                          |
| []        | null                                          |
|           | null                                          |
+-----------+-----------------------------------------------+

whereas for example in duckDb you get :

┌───────────┬──────────────────────────────────────────────────────────┐
│    bar    │ COALESCE(CAST(list_contains(bar, 1) AS VARCHAR), 'null') │
│  int32[]  │                         varchar                          │
├───────────┼──────────────────────────────────────────────────────────┤
│ [1, 3, 5] │ true                                                     │
│ []        │ false                                                    │
│           │ null                                                     │
└───────────┴──────────────────────────────────────────────────────────┘

Expected behavior

array_has([], 1) => false.
array_has(null, 1) => null.

Additional context

According to me, the issue is situated in the array_has_dispatch_for_scalar method :

...
for (i, offset) in offsets.windows(2).enumerate() {
        let start = offset[0].to_usize().unwrap();
        let end = offset[1].to_usize().unwrap();
        let length = end - start;
        // For non-nested list, length is 0 for null
        if length == 0 {
            continue;
        }
        let sliced_array = eq_array.slice(start, length);
        final_contained[i] = Some(sliced_array.true_count() > 0);
    }
...

There is no distinction between null or length = 0.
You cannot only look at the offset difference. You have to look at the validity buffer as well.

[alamb]

I agree the expected behavior listed here makes more sense than the current datafusion behavior