The agent section in Declarative Pipelines should be pluggable, so that we can have multiple implementations with their own logic. Once JENKINS-37011 is done, we'll be able to implement steps that compose other steps, which is what we're going for here.

So we'll have, e.g., a "label" provider for just running on an agent directly, a "docker" provider for Docker Pipeline, a "kubernetes" provider for the Kubernetes plugin, a "docker-slaves" provider for the Docker Slaves plugin, etc.