Support parallelization of child partition .close() operations

[cwensel]

By default, we don't leverage threading server/processing side since it's a highly constrained resource.

But in local mode, and when partitioning sink data into multiple files, parallelizing the close operation makes sense as the purges may include more than one partition.

Providing hooks to wrap the cache of open partitions purge and close operations would benefit applications that were designed specifically for partitioning data if the container hosting the application can provide cpus for the threads.