Update documentation to be clearer about where to add auto injection annotations

Nobody has claimed this yet.

When I was setting linkerd2 up for the first time I added my `inkerd.io/inject: enabled` annotation to the deployment, and I was seeing the injector skipping it. After discussing on slack I was pointed in the right direction and moved it pod template where things worked correctly.

However, the documentation in a couple of places seems to suggest that my original way would have worked correctly. The wording which was throwing me off was:

>Linkerd automatically adds the data plane proxy to pods when the `linkerd.io/inject: enabled` annotation is present on a namespace, deployment, or pod.

from here: https://linkerd.io/2/features/proxy-injection/

>This is typically done by annotating the namespace, deployment, or pod with the `linkerd.io/inject: enabled` Kubernetes annotation, which will trigger automatic proxy injection when the resources are created.

from here: https://linkerd.io/2/tasks/adding-your-service/

Start with the proxy-injection and adding-your-service documentation pages linked in the issue. Check how the annotation is described for namespaces, deployments, and pod templates, then make the wording unambiguous about where it must be placed and verify both pages consistently explain the working setup.

Written by the indexing model from the issue text.