Diagnose CI/CD queue bottlenecks and propose three throughput improvements

This tests systems thinking on CI/CD scheduling and queuing theory beyond tool familiarity. A strong answer first splits queue wait from job execution via platform metrics, then proposes three fixes: right-sizing parallelism to cut wall-clock time, aggressive caching and artifact reuse to remove redundant work, and workload sharding or dynamic agent scaling to absorb spikes. A red flag is immediately adding build agents without first measuring whether the bottleneck is queue depth, slow tests, or serial stages.