Implement streaming approach for DeviceSelect and DevicePartition

[jrhemstad]

PR #2205 was a prototypical implementation of a streaming implementation for DeviceSelect, which was good enough to gather some performance data but had several limitation:

• the number of selected items actually was not written to a temporary to propagate to the i+1-th invocation

• DevicePartition does not work out of the box, because the rejected items are written in reverse order from the end of the given output iterator. This cannot trivially be implemented using a fancy iterator approach as originally used in [DRAFT]: Experimental: Streaming DeviceSelect #2205

• Implement [FEA]: Tabulate Output Iterator #1486 to redirect num_selected_out to either a temporary or the user-provided num_selected_out, depending on whether it's the last invocation

• Add a kernel parameter that captures the streaming state of previous invocations

• Adapt AgenetSelectIf to incorporate the streaming state