Use faster PRNG in the allocations profiler

[kpamnany]

rand() locks and is slow. This uses the seed from ptls.

[NHDaly]

Before and after evidence of the improvement:

6 threads:

julia> @time let out = zeros(Int, 1000)
           Threads.@threads for i in 1:1000
               out[i] += sum(Any[rand(Int) for i in 1:300000])
           end
           sum(out)
       end
  1.987923 seconds (600.03 M allocations: 11.178 GiB, 42.80% gc time, 4.52% compilation time)
-7843380098545669747

julia> using Profile

julia> @time Profile.Allocs.@profile sample_rate=0.000001 begin
           let out = zeros(Int, 1000)
               Threads.@threads for i in 1:1000
                   out[i] += sum(Any[rand(Int) for i in 1:300000])
               end
               sum(out)
           end
       end
  8.568267 seconds (600.03 M allocations: 11.178 GiB, 9.67% gc time, 0.86% compilation time)
3995634343101539424

shows a significant difference.

After this PR, the difference is gone! 🎉

julia> @time let out = zeros(Int, 1000)
           Threads.@threads for i in 1:1000
               out[i] += sum(Any[rand(Int) for i in 1:300000])
           end
           sum(out)
       end
  1.934198 seconds (600.08 M allocations: 11.181 GiB, 43.48% gc time, 12.11% compilation time)
6985800949357430475

julia> using Profile

julia> @time Profile.Allocs.@profile sample_rate=0.000001 begin
           let out = zeros(Int, 1000)
               Threads.@threads for i in 1:1000
                   out[i] += sum(Any[rand(Int) for i in 1:300000])
               end
               sum(out)
           end
       end
  2.157459 seconds (600.03 M allocations: 11.178 GiB, 39.03% gc time, 3.48% compilation time)
6035867019842016171