Multi-token streaming rolling out, delivering tokens in batches at 200 events per second

We're gradually introducing multi-token streaming across our models. Instead of sending tokens individually, we'll now deliver them in batches through 200 evenly-spaced events per second. This change eliminates the artificial delays that occurred with single-token streaming. Please verify that your application doesn't depend on specific token delivery patterns.