Long stalls when using extended sessions with new partition leasing mechanism

The new partition lease mechanism of the AzureStorage provider waits for sessions to complete before releasing the ownership lease.

If extended sessions are enabled, this can cause unexpected long delays during which the partition is effectively stalled.

As a solution to this problem, I think we should cancel the extended sessions when partitions are shutting down.

Also, it may make sense to address this a bit more generally and cancellation tokens to `TaskActivityWorkItem` and `TaskOrchestrationWorkItem`. This will allow the backend to communicate to the front end when the partition is shutting down and that it would be better to cancel this work item (if it has not already started). For example, this would allow large entity operation batches to terminate sooner.