Is it common to manage environments for different projects that use the same pipeline? Even when using an orchestrator like Kubeflow ? Or is it a better alternative to create a pipeline class and then register different pipelines that would inherit it and change the datasets, keeping the environments in the canonical form.