MultiVectorRetriever : could it be more useful?

[magaton]

Currently, MultiVectorRetriever is created as a wrapper around vector store and doc store. And that is fine, when we have only dense vector search based retrieval.
However, the EnsembleRetriever allows combining outputs of several retrievers, e.g Dense vector, Full Text Search and/or Sparse Vectors.

In the ingestion time (I am using langchain's indexing API) the chunks are indexed into langchain_pg_embedding table. I have expanded the table so that chunk content is indexed there as well.

So, if the same content (chunk) is indexed in the same table (via embedding and fts_index fields), then we need MultiVectorRetriever to operate on Retriever, not Vector Store level.

Question here is:
Shouldn't MultiVectorRetriever accept BaseRetriever instead of VectorStore, as the arg?