dicts: improve performance for non-concrete values and keys

[KristofferC]

Same as JuliaCollections/OrderedCollections.jl#166

Benchmark:

  using DataStructures
  using BenchmarkTools

  const N = 1000
  pairs = [i => (iseven(i) ? i : Float64(i)) for i in 1:N]

  d_base  = Dict{Int,Number}(pairs)
  d_robin = RobinDict{Int,Number}(pairs)
  d_ord   = OrderedRobinDict{Int,Number}(pairs)
  d_swiss = SwissDict{Int,Number}(pairs)

  function sumvals(d)
      s = 0.0
      for (k, v) in d
          s += v
      end
      return s
  end

  print("Dict (base):       "); @btime sumvals($d_base)
  print("RobinDict:         "); @btime sumvals($d_robin)
  print("OrderedRobinDict:  "); @btime sumvals($d_ord)
  print("SwissDict:         "); @btime sumvals($d_swiss)

Results:

  Dict (base):         15.870 μs (1000 allocations: 15.62 KiB)

  Before
  OrderedRobinDict:    145.245 μs (3234 allocations: 66.16 KiB)
  SwissDict:           142.385 μs (3234 allocations: 66.16 KiB)

  After
  OrderedRobinDict:    13.913 μs (1000 allocations: 15.62 KiB)
  SwissDict:           13.795 μs (1000 allocations: 15.62 KiB)