Removes redundant division from avg() function in case of parallel non-keyed GROUP BY, e.g.

SELECT AVG(price) FROM trades;

Native avg*Acc() functions were using avg += (v - avg) / c in each iteration while the outer Avg*VectorAggregateFunction Java function was multiplying the average and the count before storing them in the partial result (sum.add(value * count)).

Benchmarks

Test box: Ryzen 7900x, 64GB RAM, Ubuntu 24.04

CREATE TABLE IF NOT EXISTS 'x' ( 
        d DOUBLE,
        ts TIMESTAMP
) TIMESTAMP(ts) PARTITION BY DAY;

INSERT INTO x SELECT
    rnd_double() * 20 + 10 AS d,
    generate_series as ts
  FROM generate_series('2025-01-01', '2025-01-02', '172u');

-- before: 90ms
-- after: 74ms
SELECT AVG(d) FROM x;

With ClickBench's hits table the difference is:

-- before: 19ms
-- after: 15ms
SELECT AVG(UserID) FROM hits;