Clones for parallel execution would share a common structure handing off partial ranges (>> DOP) of the collection, that each stage object would consume before asking for another one

Includes unit tests for both forward and backward scanning, and performance tests to verify the speedup over the DOP=1 scenario 

As parallel queries create extra Client objects that could interfere with the normal Client creation performance, create a test that measures the rate of handling new connection requests in both scenarios (single-threaded and parallel queries) and if the rate drops, see whether the parallel threads in the pool can keep their Client object alive across usages