improve performance of sync* code on VM

[mraleph]

• Do not use Closure.clone to create clones of the :sync_body closure - instead change kernel transformation to put :sync_body into another closure and call that closure to produce a copy of :sync_body. Prototype change improves performance in JIT mode by 3x.
• Disable argument type checks for sync-yielding functions because their arguments are guaranteed to match.
• In _SyncIterator recognize yield* of a _SyncIterable and track the top of the iterator stack explicitly rather than delegating downwards which makes deep recursive yield* ineffecient. (Apparently dart2js does that which makes it more efficient than VM /cc @rakudrama ).

Related issue(PR) on Flutter side (flutter/flutter#36303). /cc @jonahwilliams

/fyi @mkustermann

@mraleph do you have a benchmark tracking what we're hoping to improve for the third point?

[mraleph]

There are benchmarks on Golem which exercise this. Check for example Iteration/dart/TreeIteration.dart it has recursive sync* iterable which would benefit from this improvement.

Thanks, @mraleph. I was able to find the benchmarks in the Golem repo.

Now, for the second point on the list, can you provide a bit more context? I take it you have a specific change in mind for that one?