PP-2117: When Kelvin is executing a streaming plan, don't call the batch API when the query completes.

Summary:
Currently, there are two APIs for Kelvin sending results to the query broker: TransferResultChunk and ReceiveAgentQueryResult. TransferResultChunk is streaming and will subsume ReceiveAgentQueryResult, which is batch. The query broker has changes to receive results from both, but this Kelvin change is needed so that we only call the ReceiveAgentQueryResult API in situations where the Kelvin plan produces a memory sink.

ReceiveAgentQueryResult reads memory sinks, but in the new streaming API, the sinks will send the data directly to the query broker as GRPC sinks. As a result, during the temporary phase where we support both APIs, we differentiate between which API to use to send the results based on whether or not there are memory sinks in the final plan.

Afterwards, ReceiveAgentQueryResult will be fully deprecated and a lot of this code will go away.

Also, deprecated query_str fully from the execute query request, because it is no longer used anywhere.

Test Plan: added unit test

Reviewers: zasgar, michelle, philkuz, #engineering

Reviewed By: zasgar, #engineering

JIRA Issues: PP-2117

Differential Revision: https://phab.corp.pixielabs.ai/D5961

GitOrigin-RevId: ae680d647a895adbc965e8b2b36c693534bb9656