Have there been thorough tests to ensure that everything in the Go channel is completely flushed when the OpenTelemetry collector service shuts down? We use a batch -> ClickHouse pipeline to store logs. We auto-scale the pods via HPA.