Document per-workload / ns annotation of shutdown enabling

#1816 did not capture the fact that you can enable this per workload or possibly per-ns.

From slack:
> config.linkerd.io/proxy-admin-shutdown: enabled can be set on pod specs and, i believe, namespaces (would need to confirm that it works at the namespace level) to opt-in to the old behavior

Start by reviewing issue #1816 and the existing documentation for config.linkerd.io/proxy-admin-shutdown. Verify whether the annotation works on pod specs and namespaces, then document the supported per-workload and per-namespace usage and the old behavior it enables.

Written by the indexing model from the issue text.