[nnbd/corelib] Fix performance regression in _SyncIterator.moveNext

In Iteration.deeptree.syncstar benchmark _SyncIterator.moveNext takes
majority of time. If _moveNextFn is cached in a local variable,
then loading a function object out of _moveNextFn closure is hoisted
out of the loop and executed even if _moveNextFn is not called.
The function and cached moveNextFn are also saved to frame.
This extra code results in a slight slowdown of _SyncIterator.moveNext
when _yieldEachIterator != null.

Iteration.deeptree.syncstar(RunTime) 98877 -> 81541 us
(with NNBD core libraries)

Change-Id: Ieac94e106632f9c923eb72d9221bcdecb52fa652
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/138609
Reviewed-by: Martin Kustermann <[email protected]>
Commit-Queue: Alexander Markov <[email protected]>

Author: alexmarkov

sdk_nnbd/lib/_internal/vm/lib/core_patch.dart

@@ -136,8 +136,7 @@ class _SyncIterator<T> implements Iterator<T> {
   _SyncIterator(this._moveNextFn);
 
   bool moveNext() {
-    final moveNextFn = _moveNextFn;
-    if (moveNextFn == null) {
+    if (_moveNextFn == null) {
       return false;
     }
     while (true) {
@@ -150,7 +149,7 @@ class _SyncIterator<T> implements Iterator<T> {
       }
       // _moveNextFn() will update the values of _yieldEachIterable
       //  and _current.
-      if (!moveNextFn(this)) {
+      if (!_moveNextFn!(this)) {
         _moveNextFn = null;
         _current = null;
         return false;