feat(deepresearch): Introduce HyDE into the RAG process

[benym]

Describe what this PR does / why we need it

Introduce the transformation process of Hypothetical Document Embeddings(HyDE) into the existing deepresearch hybrid rag processing process

HyDE paper is https://arxiv.org/abs/2212.10496

As a processing method in rag, it has been integrated in langchain

Does this pull request fix one issue?

Describe how you did it

Describe how to verify it

Special notes for reviews

[benym]

ptal,thx @GTyingzi

[benym]

Hello, do you have time to review this pr and #17 @VLSMB

[VLSMB]

Hello, this suggestion of yours feels excellent. Could you share a test screenshot to demonstrate the effect?

[benym]

Hello, this suggestion of yours feels excellent. Could you share a test screenshot to demonstrate the effect?

ok, when using HyDE, the model will generate a hypothetical answer based on the user's question, as follows

This will then be provided as a query for the subsequent processes, which can effectively increase the recall rate

The following picture is from the original paper

Experiments on multiple datasets show that the Normalized Discounted Cumulative Gain(NDCG) and Recall metrics have been

significantly improved, more experiments details can be seen in the paper

[VLSMB]

LGTM, thx for your contribution!