Add block param for benchmarks to allow them to do their own looping.

This helps in the case that a benchmark had some expensive setup. Previously,
* The setup could be done once, outside the benchmark:

  expensive_setup()
  tach("method1") do
    # code to benchmark
  end

* The setup could be done every test, skewing the benchmark:

  tach("method2") do
    expensive_setup()
    # code to benchmark
  end

With this change, you have a third option.

* The setup is done once, and the benchmark loops itself:
  tach("method3") do |n|
    expensive_setup()
    n.times do
      # code to benchmark
    end
  end

One unfortunate side-effect of this is that using the new syntax will result in
lower benchmark scores because of avoiding the overhead of calling a block
multiple times.

lib/tach.rb

@@ -54,9 +54,13 @@ def run_tach(name, count, &benchmark)
       print(name)
       STDOUT.flush
       tach_start = Time.now
-      for index in 1..count
-        instance_eval(&benchmark)
+
+      if benchmark.arity == 0
+        count.times { benchmark.call }
+      else
+        benchmark.call(count)
       end
+
       tach_finish = Time.now
       duration = tach_finish.to_f - tach_start.to_f
       duration
@@ -90,6 +94,12 @@ def run_tach(name, count, &benchmark)
       [$1, $2]
     end
 
+    tach('regex 4') do |n|
+      n.times do
+        header = data.match(/(.*):\s(.*)/)
+        [$1, $2]
+      end
+    end
   end
 
 require 'benchmark'