A financial services company uses an Amazon Bedrock Knowledge Base backed by an OpenSearch Serverless vector collection. During daily ingestion of updated regulatory filings, they observe that newly synced documents are searchable within seconds, but query latency for the RAG application spikes significantly during the ingestion window, causing timeouts in their customer-facing chatbot. They want to eliminate the latency spikes during ingestion while keeping the solution fully managed and cost-effective for a workload with predictable, bursty ingestion once per day. Which approach best addresses this problem?