improve performance of `rand(n:m)`

[jlapeyre]

A recently published algorithm is unbiased and more performant than rand(n:m).

See https://discourse.julialang.org/t/rand-1-10-vs-int-round-10-rand/14339/10.

[affans]

Reference issue on the speed of rand() #20582
Maybe the PR there can be updated to use this algorithm.

[rfourquet]

Closed by #29240.