Offload filter evaluation of incoming recordbatches with selection vectors

[bobluppes]

At the moment, only incoming recordbatches without selection vectors are offloaded to cpp code in the accelerated filter operator.

Support for recordbatches with existing SV2 or SV4 vectors could be added by passing the incoming selection vector as an input buffer. The kernel should then only evaluate the filter for records in the batch coinciding with the indices in the selection vector.