quite faster rand(::MersenneTwister, ::Type{Float64}) etc... (#37916)

Generation of `Int64` just became faster than of `Float64` (#37914).
Given that dSFMT produces natively `Float64`, and that generating a random
`Float64` needs only 52 random bits, surely performance can be improved there.

Just ask the `gen_rand` routine, which randomizes the cache for floats,
to not inline, as it's unlikely to be called (once in a thousand).

The speedups are about 1.8x for `rand(rng, Float64)`, 13% for
`randn(rng, Float64)`, more than 1.9x for `rand(rng, Int32)`
and smaller integers (including `Bool`).

Author: rfourquet

stdlib/Random/src/RNGs.jl

@@ -224,7 +224,7 @@ mt_setfull!(r::MersenneTwister) = r.idxF = 0
 mt_setempty!(r::MersenneTwister) = r.idxF = MT_CACHE_F
 mt_pop!(r::MersenneTwister) = @inbounds return r.vals[r.idxF+=1]
 
-function gen_rand(r::MersenneTwister)
+@noinline function gen_rand(r::MersenneTwister)
     r.adv_vals = r.adv
     GC.@preserve r fill_array!(r, pointer(r.vals), length(r.vals), CloseOpen12())
     mt_setfull!(r)