bug(udf): incorrect error handling of non-batched UDF

[fuyufjh]

Describe the bug

For non-batched UDFs, the users expect the function to be executed row by row. If exception is thrown, the corresponding row should get a NULL result.

However, due to the current implementation of UDF framework, the actual behavior is: once a function failed, the whole batch will fail and NULLs will be filled for all these rows.

Even worse, this actually produce inconsistent data to downstream, because the behavior depends on how the input data was chunked.

Error message/log

NA

To Reproduce

CREATE FUNCTION divide(n INTEGER) RETURNS FLOAT 
LANGUAGE JAVASCRIPT AS $$
    export function divide(n) {
        if (n === 0) {
            throw new Error("Division by zero!");
        }
        return 100 / n;
    }
$$;

CREATE TABLE input (
    value INTEGER
);

INSERT INTO input SELECT generate_series(-10, 10);

CREATE MATERIALIZED VIEW result AS
SELECT 
    value,
    divide(value) as result
FROM input

SELECT * from result;

You will see multiple results are null.

Another problem is delete from input doesn't work... Didn't investigate yet, perhaps because of the data inconsistency.

Expected behavior

Only one row is null in the result.

How did you deploy RisingWave?

No response

The version of RisingWave

No response

Additional context

No response