You're right that there are some metrics of diversity that improve research quality. For example, interdisciplinarity, where you're getting diverse expertise joining together to solve a problem, does lead to better outcomes. Some of the biggest and most important breakthroughs in science come at the intersection of disciplines. Interdisciplinary diversity does improve science.
Whether or not I have people on my team who are exclusively homosexual or heterosexual, or self-identify as Latinx does not help me solve the distribution of prime numbers. It almost seems laughable that in the 21st century I would have to make that point.
Intellectual diversity does improve science. All of the other metrics of diversity don't.