In POA's router, we're trying to share the collection properties across batches to determine how operations are routed to which worker threads (the hash depends on collection properties such as collation, etc).

The POA router allows this cache to live until a DDL operation is seen, at which point we drain all in-flight work, and then reset the cache, before continuing.

After SERVER-136466, we will be resetting the cache every batch. There shouldn't be a correctness risk because DDL operations are still applied serially and all prior operations are drained first, so there should be no risk of a new cache being used while operations utilizing the former cache are in progress (this would result in different worker threads operating on the same document).

However, because we're resetting the cache every batch after SERVER-136466, there might be a performance hit. It should be negligible, but let's double check.