Support text/image multi-modal embeddings (start with OpenClip)

[richard-epsilla]

Describe the feature

• Support OpenClip embedding to unify text and image vector embedding space, and support cross modality search

Motivation and use case

• Multi-modal search

Additional context

• to be filled