Optimize range iteration

For loops support a range syntax for iterating a fixed number of times,
which looks like this.

```liquid
{% for i in range (1..100) %}
        ...
{% endfor %}
```

Previously, we converted these ranges to arrays using `to_a`, which
initialized an array containing each number in the range. Since all we
use these ranges for is iteration, this is far less efficient than
using a range iterator.

Doing this means that iterating over ranges now takes O(1) rather than O(n)
memory. See the PR for more benchmarks.

* Remove to_a cast on ranges
* Add ReversableRange iterator
* Add custom range-specific slicing logic

test/integration/tags/for_tag_test.rb

@@ -36,6 +36,10 @@ HERE
     assert_template_result('321', '{%for item in array reversed %}{{item}}{%endfor%}', assigns)
   end
 
+  def test_for_range_reversed
+    assert_template_result('321', '{%for item in (1..3) reversed %}{{item}}{%endfor%}', {})
+  end
+
   def test_for_with_range
     assert_template_result(' 1  2  3 ', '{%for item in (1..3) %} {{item}} {%endfor%}')