Flushing in the span batcher is currently global: if a flush event is emitted (e.g. because a bucket became full), we flush all buckets. Ideally, all the limits (size, weight, time) should be per-bucket, and we should only flush that specific bucket instead of the whole buffer.