Consider parallelization of gets in source/mv

[purplefox]

When a new rows arrives on a source or MV often the previous value for the row needs to be loaded from storage. Performing a get for every incoming row can create a bottleneck in the ingestion as the latency for a get can be significant. We can consider performing all or some of the gets for a batch in parallel in order to reduce the overall get latency for the batch.