separate steps for generating rootfs

[deitch]

This does not change anything in the output, only in the interim steps.

Current

Currently, we generate the rootfs.img by calling makerootfs.sh, which calls (simplified):

linuxkit build -o - $YML | docker run mkrootfs-squash

In other words, we build the OS root filesystem (left hand side of above) and pipe the tar stream to the container that creates squashfs (or ext4).

This works, but has two shortcomings:

1. If we need to regenerate, we have to do the whole thing, both expensive parts.
2. We cannot use the interim output tar for anything useful (like, perhaps, scanning for an SBoM or security)

Proposed

This PR breaks the above down into 2 steps:

1. Generate the rootfs.tar file
2. Use the tar file to generate the rootfs image (squashfs or ext4)

The rootfs.tar is in the dist/<arch>/<release>/ directory, but not in the installer/ subdir, as we do not need the tar file artifact in the final eve image.

A next step PR will add scanners, initial for an SBoM which will be included in installer/ and thus in the resulting image, but possibly also in other places (release artifacts), and future security scanners.

How

Change:

• tools/makerootfs.sh to have 3 different operating modes, as well as use getopt to parse the CLI flags
• Makefile to add a rootfstar target and several target changes
• docs/BUILD.md to describe the new usage

[deitch]

I fixed most of the yetus complaints. All of the remaining yetus complaints are incorrect here.

[eriknordmark]

LGTM