We have memory errors in the IndexBuildsCoordinator's use of the CollectionIndexBuildsTracker and DatabaseIndexBuildsTracker objects where we call waitUntilNoIndexBuildsRemain through hash map iterators rather than acquiring a safe shared_ptr to call wait upon, leading to a use after free when the entry is removed from the hash map and the caller of waitUntilNoIndexBuildsRemain does not have a shared_ptr instance to keep the object in memory.