This has already been detailed as a GitHub issue, but it was closed without much consideration to the benefits of having the feature implemented server-side:
I’ve also reached a point where I’ll be doing a lot of random offsets in my queries, and doing two (or more, with nesting) round-trips to properly handle node count becomes a pretty chunky performance cost. Further, because offset is denoted in pages, this has a tight bound on randomness for >1 elements: if I want 4 random elements, they must be either independently queried or batched as a page.
Would really love to see this implemented