Add `TopK`

[mcabbott]

This wants to add a way to keep the largest k elements.

Many ways to do this were recently timed here: https://discourse.julialang.org/t/maintaining-a-fixed-size-top-n-values-list/78868/15 This matches the simplest implementation there. It's possible that the slight gains of the others are worth it, although on my computer they don't seem to be visible, on that example.

However, the mutable struct with o.n += 1 at every step seems to make a huge difference. Is this expected, and or is there a way around it?

julia> v1 = init1(N); v1 = min_values1(N, x, v1); v11 = init1(N);

julia> @btime min_values1($N, $x, $v11);
  10.270 ms (0 allocations: 0 bytes)

julia> v4 = init4(N); v4 = min_values4(N, x, v4); @assert last(v1) == last(v4); v44 = init4(N);

julia> @btime min_values4($N, $x, $v44);
  10.370 ms (0 allocations: 0 bytes)

julia> v6 = init6(N); v6 = min_values6(N, x, v6);

julia> @btime min_values6($N, $x, $v66);
  10.248 ms (0 allocations: 0 bytes)

# This PR:

julia> v7 = fit!(TopK(N; rev=true), x).value;

julia> @assert sort(v7) == sort(v1)

julia> @btime fit!($(TopK(N; rev=true)), $x);
  33.560 ms (0 allocations: 0 bytes)
  15.436 ms (0 allocations: 0 bytes)  # without o.n += 1 step
  10.273 ms (0 allocations: 0 bytes)  # with immutable struct, and without o.n += 1