enhance: run retrievers within the merging retriever concurrently

[iwilltry42]

For a medium-sized heterogeneous dataset, this shaves off 2s from retrieval time (before: ~8s)