Design item expiry processing with concurrent compaction with non-blocking writes

With couchstore, either doc writes or compaction runs at a time. Writes are blocked while compaction is running. This makes expiry processing simple during the compaction. Any race condition of an expired item getting overwritten while processing expiry during compaction can be avoided by checking the latest hash table entry. When concurrent compaction is executed, we have to make sure that a new version of the item has been already persisted in the storage engine. We need to design the expiry processing APIs with concurrent compaction.