[CFE] Experiment with benchmarking (part 2)

For a 3-iteration run of compiling dart2js in benchmark mode run like
```
pkg/front_end/tool/fasta -Dbenchmark=true -Diterations=3 compile pkg/compiler/bin/dart2js.dart
```

the 3rd iteration took 7352 ms, the majority being spent on

790 ms for outline
102 ms for "buildComponent"
283 ms for buildClassHierarchyMembers
152 ms for buildOutlineExpressions (of which ~2ms is spent in inferer)
3764 ms for buildBodies (of which
  * ~159 ms is spent tokenizing,
  * ~52 ms spent (emulated) diet parsing,
  * ~166 ms spent (emulated) parsing,
    (we assume thus that ~166 ms is spent really parsing as well),
  * ~1255 ms is spent inferering).
662 ms for runBuildTransformations
849 ms for writeComponent
244 ms for empty visit of ast

Note that this adds a few more parsings (emulated diet parsing
and emulated parsing) to give an indication of how much time is
spent parsing, as well as an additional visit of the output ast
for reference when looking at for instance inference time and
serialization time.

Change-Id: I59aeb4cd5add0d84ed6f3a9313ea90bd67cb8816
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/233302
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Commit-Queue: Jens Johansen <jensj@google.com>