Implement async get_relevant_documents

[lukacerr]

.ainvoke requires async _aget_relevant_documents implementation to work properly.

Exhibit:

Traceback (most recent call last):
  File "/home/luka/Documents/nexata/llm/.venv/lib/python3.13/site-packages/grpclib/server.py", line 445, in request_handler
    await method_func(stream)
  File "/home/luka/Documents/nexata/llm/lib/nexata/__init__.py", line 54, in __rpc_health
    response = await self.health(request)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/luka/Documents/nexata/llm/src/services/health.py", line 28, in health
    doc = await ret.ainvoke("das")
          ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/luka/Documents/nexata/llm/.venv/lib/python3.13/site-packages/langchain_core/retrievers.py", line 326, in ainvoke
    result = await self._aget_relevant_documents(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        input, run_manager=run_manager, **kwargs_
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/home/luka/Documents/nexata/llm/.venv/lib/python3.13/site-packages/langchain_core/vectorstores/base.py", line 1119, in _aget_relevant_documents
    raise ValueError(msg)
ValueError: search_type of vector not allowed.

Code that replicates the given error:

# Vector store initialization
vstore = AzureCosmosDBNoSqlVectorSearch(
    search_type="vector",
    database_name="nexata",
    container_name="vectors",
    cosmos_client=CosmosClient(
        getenv("AZURE_COSMOS_DB_ENDPOINT", "https://127.0.0.1:8081"),
        getenv("AZURE_COSMOS_DB_KEY", ""),
    ),
    embedding=embeddings,
    vector_embedding_policy={
        "vectorEmbeddings": [
            {
                "path": "/embedding",
                "dataType": "float32",
                "distanceFunction": "cosine",
                "dimensions": 1536,
            }
        ]
    },
    indexing_policy={
        "automatic": True,
        "indexingMode": "consistent",
        "includedPaths": [{"path": "/*"}],
        "excludedPaths": [{"path": '/"_etag"/?'}],
        "vectorIndexes": [{"path": "/embedding", "type": "diskANN"}],
        "fullTextIndexes": [{"path": "/text"}],
    },
    cosmos_container_properties={
        "partition_key": PartitionKey("/metadata/tenant_id"),
    },
    cosmos_database_properties={},
    vector_search_fields={
        "text_field": "text",
        "embedding_field": "embedding",
    },
    full_text_policy={
        "defaultLanguage": "en-US",
        "fullTextPaths": [
            {"path": "/text", "language": "en-US"},
        ],
    },
    full_text_search_enabled=True,
)

ret = vstore.as_retriever(search_type="vector")

# Later, in an async function...
doc = await ret.ainvoke("das")  # ValueError: search_type of vector not allowed.

[marlenezw]

Hi @aayush3011, do you have any bandwidth to review this? It's a super helpful addition. Linting looks like it's failing because of a signature mismatch.

[marlenezw]

@aayush3011 I'm wanting to merge or close all PR's in the coming days. Please take a look