This is done by running `setcap cap_net_bind_service=+ep` on the executable
in the build stage (doing it in the runtime stage creates an extra copy of
the executable that bloats the image). This only works when using the
BuildKit-based builder, since the `COPY` instruction doesn't copy
capabilities on the legacy builder.
This section of code seems to be breaking the Docker release workflow as of a
few days ago, though it's unclear why. This tag only existed to work around
an issue with Docker pulling the wrong image for ARMv6 platforms; that issue
was resolved in Docker 20.10.0, which has been out for a few years now, so it
seems like a reasonable time to drop this tag.
Jeremy Lin
