Optimize #rotate!

[HertzDevil]

This PR has a few related optimizations:

• Ensures that passing n = 0 or a multiple of size in the default Indexable::Mutable implementation skips the loop altogether.
• Moves the existing optimization in Array#rotate! to Slice, so that any Slice-based container could leverage it. This includes BitArray (only when n and size are both multiples of 8, i.e. the rotation can be done on byte boundaries), and StaticArray.
• Adds an optimization for BitArrays when rotating 32 or fewer elements in either direction. It uses integer shifts and continues taking O(1) space, so it is very fast compared to the default algorithm. Benchmarks:

  require "benchmark"
  require "bit_array"
  
  SIZES = [0, 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 5, 10, 20, 50, 100, 200, 500].sort!
  
  x : BitArray? = nil
  Benchmark.ips do |b|
    SIZES.each do |size|
      {8, 19}.each do |n|
        b.report("new rotate!(#{n})") do
          ba = BitArray.new(size)
          ba.fill { rand > 0.5 }
          100.times { ba.rotate!(n) }
          x = ba
        end
        # ditto for the old `#rotate!`
      end
    end
  end

  

[sdogruyol]

🚀 Thank you @HertzDevil 🙏