Adding late interaction support to the QdrantVectorStore in langchain_qdrant

[ebettenga]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

I could be wrong about how this works, It it looks to me in QdrantVectorStore that we do prefetching for sparse and dense vectors when using RetrievalMode.HYRBRID. with this setup it should be possible to implement the final example multi step query. That said, I don't think I'm getting passed a kwargs blob to expand that search to include the late interaction params. That would be super helpful for me and my research, and I would happily implement a fork if that's an option.

Motivation

My company is researching the best way of filtering query results, and I think this integration would unlock a new set of options when experimenting.

Proposal (If applicable)

potentially create a new RetrievalMode, or perhaps understand more about how I could achieve the payload below from here

{
"prefetch": {
"prefetch": {
"query": [1, 23, 45, 67], // <------ small byte vector
"using": "mrl_byte"
"limit": 1000
},
"query": [0.01, 0.45, 0.67, ...], // <-- full dense vector
"using": "full"
"limit": 100
},
"query": [ // <─┐
[0.1, 0.2, ...], // < │
[0.2, 0.1, ...], // < ├─ multi-vector
[0.8, 0.9, ...] // < │
], // <─┘
"using": "colbert",
"limit": 10
}

which I got from here