LiteLLMEmbedder should support image embeddings when used with embedding models that support image embeddings.

We previously attempted to add this capability, but LiteLLM had issues that made the multimodal embedder API completely inconsistent with the API of other models. This made it exceedingly difficult to integrate support for these models into our existing LiteLLM client class. As a temporary workaround, we introduced VertexAIMultimodelEmbedder, which also uses LiteLLM behind the scenes but only supports a single family of multimodal models.

Given the current state of LiteLLM, it appears that the upstream issues have been resolved, making it relatively straightforward to add image support to our LiteLLMEmbedder class.

Definition of Done

1. LiteLLMEmbedder should support embedding images via the embed_image method when configured with a model that supports it.
2. The image_support method in LiteLLMEmbedder should correctly indicate whether the model supports image embeddings. (Note: In cases of uncertainty, false negatives are preferable to false positives.)
3. The VertexAIMultimodelEmbedder class should be removed, and all instances of its use should be replaced by LiteLLMEmbedder configured with the vertex_ai/multimodalembedding model.