ListCopy.toList.100 benchmark is 30% slower with unforked NNBD libraries

[alexmarkov]

This regression can be reproduced with https://dart-review.googlesource.com/c/sdk/+/134808 and the following command:

dart benchmarks/ListCopy/dart/ListCopy.dart

The following changes in NNBD core libraries slowed it down:

• _GrowableList.toList:

-  List list = growable ? new _List(length) : new _List<T>(length);
+  final list = new _List<T>(length);

• List.filled is now called from ListMixin.toList and it starts appearing in the profile.

/cc @a-siva @mraleph @lrhn

[alexmarkov]

7e0139e improved performance of _GrowableList.toList both for NNBD and default core libraries. The first reason for regression is fixed.

Although the benchmark ListCopy.toList.100 is now faster than before (both with NNBD and default core libraries), with NNBD core libraries the benchmark is still ~11% slower compared to default core libraries.

[a-siva]

Is the current NNBD number in line with the old default core libraries number?

[alexmarkov]

The numbers which I measured on my machine before and after 7e0139e:

With NNBD core libraries:
ListCopy.toList.100(RunTime) 4129 -> 2376 us.

With default core libraries:
ListCopy.toList.100(RunTime) 3006 -> 2140 us.