Enable tuple traversal optimization for multiple selectors

[preciz]

There is a 3-4x speed up for 2-3 cases "Two classes" and "Three classes" and the memory usage is only 1/4 for those compared to the original.

I also ran the benchs/finder.exs benchmarks and there the "class multiple" cases show the same speed up like the benchmark below.

  html = File.read!("benchs/big.html")
  parsed = Floki.parse_document!(html)

  Benchee.run(
    %{
      "Single class" => fn ->
        Floki.find(parsed, ".class-mw-redirect")
      end,
      "Two classes" => fn ->
        Floki.find(parsed, ".class-mw-redirect, .reference")
      end,
      "Three classes" => fn ->
        Floki.find(parsed, ".class-mw-redirect, .reference, .external")
      end
    },
    time: 10,
    memory_time: 2
  )