The new design pushes the concurrency detection to the host instead. On hte host side, we look to see if we're concurrently making feature requests for the same solution checksums. If so, those concurrent features share the same PinnedSolutionInfo and Scope objects (in a ref counted fashion so that they stay alive as long as any of the requests is still in flight).

Then, on the OOP side, we just have a mapping from ScopeId to the in-flight operation to sync it. Any OOP operations that then are working on the same scope-Id (which means they're sharing the same pinned info on the host side) can then share teh same sync operation on the OOP side.

This is safe as hte host side can only clean up if all concurrent features on the same checksum cancel/finish. Only then does the pinned scope on the host side go away. But tahts' ok as all the features on the OOP side that might care are themselves cancelled/finished.

Note: @dibarbet suggested a nice simplification where we no longer use scope-ids, but instead use solution-checksums. That change will come in a follow-up PR to keep things simple here.