meta-lxatac-bsp: emmc-image: build smaller images (faster)

[hnez]

This PR does three things:

• Revert #62 by @rohieb because it did not achieve quite what we wanted
• Remove the partition end workaround that was removed in #62 (but without changing other things like adding fill=true).
• Do not place a GPT header at the "end"¹ of the image.
• Use the existing rootfs ext4 instead of generating a new one from the .tar.

Things I am not too sure about:

• [x] Should we use some templating magic instead of writing lxatac-core-image-base-lxatac.ext4 in the genimage.config?
• [x] Are all the dependencies automatically set up so that we can rely on lxatac-core-image-base-lxatac.ext4 being present. The first version of this PR used GENIMAGE_ROOTFS_IMAGE to specify this dependency. This is not quite correct. The genimage.bbclass documentation states quite clearly that this variable is intended for .tar images that should be unpacked and re-packed into a filesystem image (thx @ejoerns ). Instead I've now set up the dependencies explicitly.
• [x] Is the rootfs grown like we want it to.

The last one I can test, for the rest I would be glad about some feedback.

---

¹ Which is the end of the image but has nothing to do with the end of the device the image is written to, where the header should be.

[rohieb]

The seond one you could also check by inspecting the DEPENDS variable in the output of bitbake -e emmc-image.

Edit: Ah, the do_genimage[depends] += "${GENIMAGE_ROOTFS_IMAGE}:do_image_complete" should do the trick, as long as the do_image_complete task in lxatac-core-image-base creates the ext4 image.

[hnez]

Seeing the core and base in the image name always irks me a little bit, as it is not very core-y or base-y with all the development tools. We should define an actual minimal image in the future and derive our full image with all the tools from there. (But that's not something we should do or discuss here)

[hnez]

As it turns out the rootfs was not resized to match the partition as we wanted it to. Newer systemd versions ship a systemd-growfs-root.service that can help us here, so I've backported it for now. Hopefully we can throw away the backport very soon.