Remove parsing time from benchmark and support comparing to a baseline. (#1389)

I'm trying to make it easier to see relative improvements in the
benchmarks, so this change:

* Moves the parsing outside of the main benchmark loop. Since that's
  all analyzer, there's nothing we can do about it, and it doesn't help
  to measure it for the formatter. This means the benchmark time doesn't
  reflect the total time spent doing work, but it means a relative
  improvement in formatting time will make a larger fractional change
  in the overall benchmark time.

  The formatter's public API doesn't have any way to pass in a parsed
  CompilationUnit and format it, so to get this working, I had to
  reimplement some of the parsing and formatting behavior from the
  DartFormatter class in run.dart. I simplified it down to just the
  bare minimum needed for benchmarks. In particular, I didn't bother
  supporting parsing and formatting a single statement. Instead, I just
  converted all of the benchmarks to be complete compilation units.

* Adds support for showing benchmark output relative to some previously
  calculated baseline time. This makes it easier to see "this is 10%
  faster" without having to do math in my head.

* Calculates time with higher precision. I'm not sure how much of a
  difference it makes, but I used microseconds and another significant
  digit.

Author: munificent

benchmark/case/block.expect

@@ -1,5 +1,5 @@
 // Test statements nested within blocks.
-{
+main() {
   {
     someLongOuterExpression +
         anotherOperand;
@@ -76,4 +76,4 @@
     someLongOuterExpression +
         anotherOperand;
   }
-}
+}

benchmark/case/block.expect_short

@@ -1,5 +1,5 @@
 // Test statements nested within blocks.
-{
+main() {
   {
     someLongOuterExpression +
         anotherOperand;
@@ -76,4 +76,4 @@
     someLongOuterExpression +
         anotherOperand;
   }
-}
+}

benchmark/case/block.stmt renamed to benchmark/case/block.unit

@@ -1,6 +1,6 @@
 40 columns                              |
 // Test statements nested within blocks.
-{
+main() {
   {
     someLongOuterExpression + anotherOperand;
     someLongOuterExpression + anotherOperand;

benchmark/case/chain.expect

@@ -1,5 +1,5 @@
 // Test long split method chains.
-target
+var result = target
     .someMethod1(
       argument1,
       argument2,
@@ -99,4 +99,4 @@ target
       argument1,
       argument2,
       argument3,
-    );
+    );

benchmark/case/chain.expect_short

@@ -1,5 +1,5 @@
 // Test long split method chains.
-target
+var result = target
     .someMethod1(
         argument1, argument2, argument3)
     .someMethod2(
@@ -39,4 +39,4 @@ target
     .someMethod19(
         argument1, argument2, argument3)
     .someMethod20(argument1, argument2,
-        argument3);
+        argument3);

benchmark/case/chain.stmt renamed to benchmark/case/chain.unit

@@ -1,6 +1,6 @@
 40 columns                              |
 // Test long split method chains.
-target
+var result = target
     .someMethod1(argument1, argument2, argument3)
     .someMethod2(argument1, argument2, argument3)
     .someMethod3(argument1, argument2, argument3)

benchmark/case/collection.expect

@@ -1,5 +1,5 @@
 // Test collection literals.
-[
+var list = [
   [
     [
       listElement1,
@@ -124,4 +124,4 @@
       name2: recordField2,
     ),
   ),
-];
+];

benchmark/case/collection.expect_short

@@ -1,5 +1,5 @@
 // Test collection literals.
-[
+var list = [
   [
     [
       listElement1,
@@ -124,4 +124,4 @@
       name2: recordField2
     ),
   ),
-];
+];

benchmark/case/collection.stmt renamed to benchmark/case/collection.unit

@@ -1,6 +1,6 @@
 40 columns                              |
 // Test collection literals.
-[
+var list = [
   [
     [listElement1, listElement2, listElement3],
     {setElement1, setElement2, setElement3},

benchmark/case/collection_large.expect

@@ -1,5 +1,5 @@
 // Test collection literals.
-[
+var list = [
   [
     [
       [
@@ -508,4 +508,4 @@
       ),
     ),
   ),
-];
+];