Add Weighted Reciprocal Rank Fusion (WRRF) to Python SDK

[bambriz]

Description

This PR adds Weighted RRF to the python cosmos db sdk. Weights can be added to the end of the RRF function of a Hybrid Full Text Query. The number of weights need to match the number of components passed into the RRF Function. A positive weight will order by descending order and a negative weight will order by ascending order for that particular component/rank.

some examples of WRRF:

query = "SELECT TOP 10 c.index, c.title FROM c ORDER BY RANK RRF(FullTextScore(c.title, ['John']), FullTextScore(c.text, ['United States']), [1, 1])"
    
    results = container.query_items(query, enable_cross_partition_query=True)
    result_list = list(results)

 query = "SELECT TOP 10 c.index, c.title FROM c " \
            "ORDER BY RANK RRF(FullTextScore(c.title, ['John']), FullTextScore(c.text, ['United States']), [0.1, 0.1])"
    results = container.query_items(query, enable_cross_partition_query=True)
    result_list = list(results)
    
 query = "SELECT TOP 10 c.index, c.title FROM c " \
            "ORDER BY RANK RRF(FullTextScore(c.title, ['John']), FullTextScore(c.text, ['United States']), [-1, -1])"
    results = container.query_items(query, enable_cross_partition_query=True)
    result_list = list(results)

[Copilot]

Pull Request Overview

This PR adds support for Weighted Reciprocal Rank Fusion (WRRF) to the Python Cosmos DB SDK to allow different weights for hybrid full text search queries. Key changes include:

• New query tests validating WRRF behavior for both weighted and non‐weighted queries.
• Updates in query planning and aggregator logic (both synchronous and asynchronous) to incorporate component weights.
• Documentation and changelog updates explaining the new WRRF feature.

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
sdk/cosmos/azure-cosmos/tests/test_query_hybrid_search_async.py	Added async test cases for WRRF with various weight scenarios and error conditions.
sdk/cosmos/azure-cosmos/tests/test_query_hybrid_search.py	Added similar WRRF test cases for synchronous queries including missing weights error handling.
sdk/cosmos/azure-cosmos/azure/cosmos/documents.py	Declared a new query feature, WeightedRankFusion, to support WRRF queries.
sdk/cosmos/azure-cosmos/azure/cosmos/aio/_cosmos_client_connection_async.py sdk/cosmos/azure-cosmos/azure/cosmos/_cosmos_client_connection.py	Updated query plan string to include WeightedRankFusion.
sdk/cosmos/azure-cosmos/azure/cosmos/_execution_context/hybrid_search_aggregator.py sdk/cosmos/azure-cosmos/azure/cosmos/_execution_context/aio/hybrid_search_aggregator.py	Modified aggregator functions to accept component weights and adjust sorting based on weight direction.
sdk/cosmos/azure-cosmos/README.md	Documented the WRRF functionality, including examples and behavior of negative weights.
sdk/cosmos/azure-cosmos/CHANGELOG.md	Updated changelog to include the WRRF feature announcement.

[azure-sdk]

API change check

API changes are not detected in this pull request.

[simorenoh]

I don't think this information is something we should be sharing with users - it's an implementation detail more than an indication of anything. A user that is using weights knows that they are applying more or less importance to a given component. (ignore the suggestion below, just trying to point to the line in question)

Suggested change

	A Negative value on a weight will reverse the ordering for that score. Usually it is descending, but a negative weight will order it ascending.
	A Negative value on a weight will reverse the ordering for that score. Usually it is descending, but a negative weight will order it ascending.

[allenkim0129]

LGTM