OpenSearch custom_query without filters

[bakerada]

Describe the bug
Currently when trying to use a custom_query for the OpenSearchEmbeddingRetriever it errors out with default filters. Standard use allows optional filters and the documentation notes that the filters are optional with custom_query. However it looks like filters are required and will error out if {} or None is provided to OpenSearchEmbeddingRetriever.run()

Bug appears to be here : https://github.com/deepset-ai/haystack-core-integrations/blob/main/integrations/opensearch/src/haystack_integrations/document_stores/opensearch/document_store.py#L499

if isinstance(custom_query, dict):
  body = self._render_custom_query(
      custom_query, {"$query_embedding": query_embedding, "$filters": normalize_filters(filters)}
  )

The call of normalize_filters errors out here if default filters are provided

FilterError: 'operator' key missing in {}

Probably a better way to fix this, but I think something like this would make sense?

if isinstance(custom_query, dict):
  query_placeholders = {"$query_embedding": query_embedding}
  if filters:
      query_placeholders["$filters"] = normalize_filters(filters)
  body = self._render_custom_query(
      custom_query, query_placeholders
  )

To Reproduce
Here's a sample piece of code I ran

from haystack_integrations.document_stores.opensearch import OpenSearchDocumentStore
from opensearchpy import RequestsHttpConnection
from haystack_integrations.components.retrievers.opensearch  import OpenSearchEmbeddingRetriever

document_store = OpenSearchDocumentStore(
        hosts=[opensearch_endpoint],
        http_auth=awsauth,
        use_ssl=True,
        verify_certs=True,
        connection_class=RequestsHttpConnection,
        index=opensearch_index,
        create_index=False,
        embedding_dim=embedding_dim,
        embedding_field='embedding'
    )

retriever = OpenSearchEmbeddingRetriever(document_store=document_store)


custom_query = {
    "query": {
        "bool": {
            "must": [
                {
                    "knn": {
                        "embedding": {
                            "vector": "$query_embedding",
                            "k": 10000,
                        }
                    }
                }
            ]
        }
    },
    "collapse": {
        "field": "name"
    }
}

embedding = np.random.random(1536).tolist()

documents = retriever.run(
    query_embedding=embedding,
    filters={},
    top_k=10,
    custom_query=custom_query
)

Works as expected when I add in the $filters parameter and valid filters.

Describe your environment (please complete the following information):

• OS: macOS Sonoma 14.5
• Haystack version: 2.7.0
• Integration version: opensearch-haystack 1.1.0