Restrict the number of lines we await during hot-streak completion generation to prevent overwhelming inference providers. Based on empirical observations, we'll initially set this limit to five. We can adjust this value later according to the A/B test results.
Test plan
Updated unit tests and verified locally.