After this commit, CollectionBulkLoader::insertDocuments() (storage document/ external sorter insertion) is not called with collection cloner mutex held. As a result, we can hit data inconsistency and buffer overflow issues.

Data inconsistency can occur because this CollectionBulkLoader::insertDocuments() is not thread-safe function.
Buffer overflow can occur when the reads (returned from sync source) are much faster than the writes (document insertion). As a result, CollectionCloner::_documentsToInsert buffer can overflow and lead to OOM issue.

Note: I was able to see this buffer overflow, lead to server crash, when I tried to disable readOnce cursor on initial sync logkeeper workload.