Added Multimodal semantic search feature

[martin-gaievski]

Description

Adding support for multimodal search.

This PR covers phase 1 of the implementation that is a simplified version that supports only models with text and image (text, image or both).
Main parts involved:

• data ingestion. new ingestion processor created for getting embeddings for documents during ingestion. Interface is:

{
  "description": "An example neural search pipeline",
  "processors" : [
    {
      "text-image-embedding": {
        "model_id": "model1234567",
        "embedding": "multimodal_embedding"
        "field_map": {
            "text": "my_text_field", 
            "image": "my_image_field" 
        }
      }
    }
  ]
} 

• search query. existing neural-search query enhanced by adding keyword for base64 encoded image. New interface looks like this:

{
  "query": {
        "neural": {
              "passage_vector": { 
                  "query_text": "Hello world", 
                  "query_image": "base64forimage_1234567890",
                  "model_id": "xzy76xswsd",
                  "k": 100
                }
           }
        }
    }
}

That is done in 2.x to be in sync with ml-commons that are also doings changes, and to be align better with 2.11 release that is coming soon. It will be upported to main later.

Issues Resolved

#318
#362

Check List

• New functionality includes testing.
  • All tests pass
• New functionality has been documented.
  • New functionality has javadoc added
• Commits are signed as per the DCO using --signoff

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[codecov]

Codecov Report

Merging #359 (fd8b1fd) into 2.x (8d749b8) will increase coverage by 0.18%.
The diff coverage is 82.66%.

@@             Coverage Diff              @@
##                2.x     #359      +/-   ##
============================================
+ Coverage     80.51%   80.69%   +0.18%     
- Complexity      447      496      +49     
============================================
  Files            39       41       +2     
  Lines          1411     1554     +143     
  Branches        210      237      +27     
============================================
+ Hits           1136     1254     +118     
- Misses          187      200      +13     
- Partials         88      100      +12     

Files	Coverage Δ
...g/opensearch/neuralsearch/plugin/NeuralSearch.java	72.72% <100.00%> (+1.29%)	⬆️
...rch/neuralsearch/processor/InferenceProcessor.java	92.71% <100.00%> (ø)
...ocessor/factory/TextEmbeddingProcessorFactory.java	100.00% <100.00%> (ø)
...or/factory/TextImageEmbeddingProcessorFactory.java	100.00% <100.00%> (ø)
...earch/neuralsearch/ml/MLCommonsClientAccessor.java	94.28% <93.10%> (-0.59%)	⬇️
...nsearch/neuralsearch/query/NeuralQueryBuilder.java	88.88% <83.33%> (ø)
...lsearch/processor/TextImageEmbeddingProcessor.java	78.43% <78.43%> (ø)

[navneet1v]

@martin-gaievski Haven't reviewed the code completely. Will try to do it by tonight

[navneet1v]

@martin-gaievski overall code looks good to me. Lets make sure that we add enough tests to make the github action pass.

This file seems missing UTs. TextImageEmbeddingProcessor

[martin-gaievski]

@martin-gaievski overall code looks good to me. Lets make sure that we add enough tests to make the github action pass.

This file seems missing UTs. TextImageEmbeddingProcessor

Sure, let me work backwards from the unit test coverage report

[jmazanec15]

How are we confirming that this works with images if we dont have model? Has offline testing been done?

[martin-gaievski]

We have tested that offline by setting up one of BedRock titan models and registering it using remote connector feature in ml-commons.