Add support for Vertex AI's multimodal embeddings

[hugoleborso]

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

Hi ! First of all thanks for the amazing work on langchain.

I recently developed a tool that uses multimodal embeddings (image and text embeddings are mapped on the same vector space, very convenient for multimodal similarity search).

The only cool option I found to generate the embeddings was Vertex AI's multimodalembeddings001 model.
I really wanted to use lanchain to retrieve embeddings using this model but the feature only exists experimentally in the js version of langchain

Motivation

I think multimodal features are about to become quite big in AI and that many users could benefit from this feature.

I saw that this was already requested in a previous issue #13400 but it was closed because it was too old.

Proposal (If applicable)

I would gladly contribute, given some guidance !
In the previous issue, @beatgeek recommened :

It seems it would be an update to this - libs/langchain/langchain/embeddings/vertexai.py where currently it's only for textembedding-gecko and lacks the function for image embeddings.

Let me know what you think :)