Your ML platform team maintains a monorepo containing eight custom Kubeflow pipeline components, each with its own container image and unit tests. A single Cloud Build trigger currently rebuilds and re-pushes all eight container images and recompiles the pipeline on every commit, even when only one component's source changes. This makes CI runs slow and costly. You want to restructure the CI/CD process so that only components affected by a given commit are rebuilt and tested, while still ensuring the compiled pipeline references the correct image versions. What is the best approach?