identity: remove requirement for coreos-assembler.basearch

[cgwalters]

This is aiming to increase compatibility of zincati with systems that aren't derived from Fedora CoreOS (e.g. IoT, desktop systems, etc.)

Additionally, even in the FCOS case, when one derives a container image from FCOS, the final build isn't generated by coreos-assembler and hence the ostree commit won't have this metadata.

In the end - I can't think of a use case for even trying to have zincati managing an image for a different architecture than the one it's compiled for.

So, copy the code I wrote for oci-spec-rs which maps from the Rust architecture to the GNU one. xref
youki-dev/oci-spec-rs@52e450e

[lucab]

This is effectively a revert of #58, and I'm fine with merging this.
But if you read through the references there the idea wasn't to support cross-arch updates, but to make sure the underlying environment is sane (i.e. the underlying basearch and platform ID can be introspected through some stable API).
The fact that the commit metadata get lost and Zincati is the first thing to break is a canary in the mine.

/cc @jlebon

[cgwalters]

OK, if you want to look at it we can - here's the neat thing, with this move we can use the standard container image metadata. Specifically I can change things (in the container case) to call https://docs.rs/ostree-ext/latest/ostree_ext/container/store/fn.query_image_ref.html and we can dig from there into https://docs.rs/oci-spec/0.5.8/oci_spec/image/struct.ImageConfiguration.html#method.architecture

(However, we would then need to re-map the Go/OCI architecture back to GNU, we have Go code for that in https://github.com/coreos/stream-metadata-go/blob/main/arch/arch.go that we'd need to put somewhere in Rust)

[cgwalters]

coreos/stream-metadata-rust#34

[cgwalters]

I've been thinking about this more, and I think

make sure the underlying environment is sane (i.e. the underlying basearch and platform ID can be introspected through some stable API).

is covered by coreos/stream-metadata-rust#34

We could also instead add rpm-ostree (really libdnf)'s idea of the architecture name into rpm-ostree status --json - not as part of the booted deployment.

But, I'm still not sure that the original change was really strongly motivated; we don't have any case for cross-architecture updates, so again the architecture that we built for Rust must be the same as the RPM/GNU/Linux one - after a suitable mapping. So it's more about maintaining that suitable mapping in a sustainable way, and I do think the stream-metadata-rust PR is that.