repeat is unnecessarily slow

[simonbyrne]

julia> @time n = repeat(collect(1:20000), inner=[84], outer=[1]);
  1.439516 seconds (20.03 M allocations: 523.687 MB, 4.77% gc time)

Writing the loops manually is 100x faster.

Unfortunately I can't think of any easy way to fix the general case that doesn't involve generated functions.

[nalimilan]

Cf. #14082

[JeffBezanson]

I believe we could at least do block copy operations instead of computing the position of every element individually. Instead of constructing an array of scalar indexes, construct an array of ranges for each element to handle inner, initially constructing only one of the outer copies. Then that data just needs to be copied to handle outer. The case of inner being all 1s can be optimized as well.

[stevengj]

(Benchmark added to BaseBenchmarks.)