Performance: AISearchConversationalCatalogHelpers added extra ovhead on cs.worker thread causing performance degradation from March to May releaseDescriptionVA takes a long time to respond when the search response includes catalog citations. While querying splunk logs we see setting `is_conversational_cache` takes considerable amount of time during VA search execution.There is more than 20secs delay between BGE reranker calls and Synthesized GR calls in sys_generative_ai_log. This is observed in releases starting XP9.Steps to Reproduce a) Set up Yokohama/Znowassistb) perform search "ipad"c) Observe the time taken for VA to return a synthesized answerActual : VA takes a long time to respond when the search response includes catalog citations. While querying splunk logs we see setting `is_conversational_cache` takes considerable amount of time during VA search execution.There is more than 20secs delay between BGE reranker calls and Synthesized GR calls in sys_generative_ai_log.Expected: VA response should be faster WorkaroundIssue: VA takes a long time to respond when the search response includes catalog citations. While querying splunk logs we see setting `is_conversational_cache` takes considerable amount of time during VA search execution.There is more than 20secs delay between BGE reranker calls and Synthesized GR calls in sys_generative_ai_log.The issue is because we have added logic to support conversational catalog render for portals from within the script.As part of the defect change, we now use conversational catalog batch api added as part of the dependency story.As a workaround we have published this update sets in different scopes.The changes in the update set include:1) Update CatalogConversationLLMUtil script to include batch api changes -> This has been officially released as part of July store app version of the catalog app2) Update AISearchConversationalCatalogHelpers to call the batch api -> officially released in YP6 and Zp13) Restrict calling AISearchConversationalCatalogHelpers in Synthesized GR script for non va clients. This has been done as part of PRB1893377Related Problem: PRB1888352