Add multimodal image embedding support to OCIGenAIEmbeddings

[fede-kamel]

Extends OCIGenAIEmbeddings with image embedding support. Multimodal models like cohere.embed-v4.0 embed text and images into the same vector space, enabling cross-modal retrieval (search images with text, find text from images).

Builds on the vision utilities from #99 and #104 (data URI convention, vision.py module).

What changed:

• vision.py gets to_data_uri() as the core encoding primitive, plus IMAGE_EMBEDDING_MODELS registry. load_image() and encode_image() now delegate to it internally.
• New embeddings/image.py with ImageEmbeddingMixin adds embed_image/embed_images without bloating oci_generative_ai.py.
• oci_generative_ai.py gains input_type and output_dimensions fields and stays focused on OCI client + text embedding only.
• embed_image() / embed_images() accept file paths, raw bytes, or data URIs. Automatically sets input_type=IMAGE.
• output_dimensions lets you configure vector size (256, 512, 1024, 1536) for embed-v4.0+.

Quick example:

from langchain_oci import OCIGenAIEmbeddings

emb = OCIGenAIEmbeddings(
    model_id="cohere.embed-v4.0",
    compartment_id="ocid1.compartment.oc1..xxx",
    service_endpoint="https://inference.generativeai.us-chicago-1.oci.oraclecloud.com",
)
text_vec = emb.embed_query("a photo of a sunset")
image_vec = emb.embed_image("sunset.jpg")

Tested against all 6 available OCI embedding models (embed-v4.0, embed-english-v3.0, embed-english-light-v3.0, embed-multilingual-v3.0, embed-multilingual-light-v3.0, embed-multilingual-image-v3.0). 38 unit tests + 43 integration tests, ruff clean.

[fede-kamel]

Integration Testing Update

Ran full integration tests for the image embedding feature — all 43 tests pass.

Models tested

Text embedding models (5):

• cohere.embed-v4.0 (1536 dims)
• cohere.embed-english-v3.0 (1024 dims)
• cohere.embed-english-light-v3.0 (384 dims)
• cohere.embed-multilingual-v3.0 (1024 dims)
• cohere.embed-multilingual-light-v3.0 (384 dims)

Image embedding models (2):

• cohere.embed-v4.0 (1536 dims)
• cohere.embed-multilingual-image-v3.0 (1024 dims)

Test coverage

Test suite	Tests	Status
Image embedding (bytes, data URI, file path, batch)	5	✅
Cross-modal similarity (text vs image)	3	✅
Output dimensions (256, 512, 1024)	3	✅
Input type override (search_document, search_query, classification, clustering)	4	✅
Multi-model text embedding (dims, batch, similarity)	15	✅
Multi-model image embedding (single, batch, cross-modal)	10	✅
Image embedding models registry	3	✅
Total	43	✅ All pass

Notes

• cohere.embed-v4.0 is not yet available in eu-frankfurt-1 (404). All tests ran against us-chicago-1.
• Unit tests also pass (38/38).

[fede-kamel]

@paxiaatucsdedu Addressed your review feedback:

1. Renamed function: embed_images() → embed_image_batch() for better clarity
2. Added README examples: New section 3b with image embedding examples

Changes rebased onto latest main.

[fede-kamel]

Integration tests passed (43/43)

Tested with real OCI GenAI API in Frankfurt region:

Image Embedding Models (multimodal):

• cohere.embed-v4.0
• cohere.embed-multilingual-image-v3.0

Text Embedding Models:

• cohere.embed-v4.0
• cohere.embed-english-v3.0
• cohere.embed-english-light-v3.0
• cohere.embed-multilingual-v3.0
• cohere.embed-multilingual-light-v3.0

Confirmed working:

• embed_image() - single image from bytes, file path, data URI
• embed_image_batch() - multiple images in batch
• Cross-modal similarity (image ↔ text vectors in same space)
• Output dimensions: 256, 512, 1024, 1536
• Input types: SEARCH_DOCUMENT, SEARCH_QUERY, CLASSIFICATION, CLUSTERING