CI: bump linuxkit

[ijc]

With linuxkit/linuxkit#2811 merged update to a newer linuxkit tool which consumes far less RAM (which should reduce/remove the occasional spurious we see here today due to the process getting OOM killed) and switch to building ISO images. Currently these are put into the workspace but not yet published as artefacts, need to think about the latter since they are quite large.

[ijc]

Comparing latest master CI to this PR the image build seems to have gone from ~2.5 mins for cri images and to ~11 mins for cri images (split ~8 mins for build and ~3 for adding to workspace) and ~20 mins (~13build + ~3 workspace) for docker ones, which is quite a slow down. I'm going to try instrumenting a bit.

Adding to the workspace seems pointless just now since nothing consumes it but one day I hope to publish and/or do a test boot which would make it useful.

[ijc]

Breakdown from https://circleci.com/workflow-run/d7b2de57-6f95-41f3-9a7b-dce9e6980886:

Runtime	Network	Import pkgs	tar	iso-bios	iso-efi	Builds (total)	Save	Job total
cri-containerd	bridge	0:16	1:25	3:32	4:00	8:59	1:57	11:55
cri-containerd	weave	0:07	1:29	3:34	3:33	8:37	2:20	11:38
docker	bridge	0:20	2:31	7:17	6:53	16:42	4:15	22:30
docker	weave	0:23	1:48	6:21	6:24	14:34	3:54	19:53

(totals aren't exact, there are other steps but they should be pretty similar).

Seems like the right thing to do is to go back to just building tar balls (but with the newer version of the tool) and consider what else we might need to build when we have a way to boot something. IOW rewind this PR back to just the first commit.

[ijc]

Backed out to just the linuxkit binary bump, should stop the occasional spurious CI failures.

[ijc]

Since the linuxkit binary likely needs updating for stuff merged over xmas I'm going to do a new PR which just does the linuxkit binary bump.