The intended fix here is the third commit: @pwhalen wants us to update the 'rawhide' tag in registries for IoT composes, but they are not actually versioned "Rawhide" like regular Fedora Rawhide composes are, they are versioned with the release number currently associated with Rawhide (e.g. 43 right now). This adds a condition to update the 'rawhide' tag if the release number of the compose being processed is the Rawhide release number.

The other two commits are test suite fixups.

The first is a necessary precursor to be able to test the intended fix here - the container tests were broken because fedfind behaviour changed so their cassettes were no longer valid, but we could not re-record the cassettes because the composes we used in those tests no longer exist on the server. So we have to change to using newer composes, change the message fixtures, update the test expectations, and re-record the cassettes to make those tests work again. Some other tests use the same fixtures, so we have to update those too. I kept the old RC compose message under a different name for the GCP tests, as I do not have API access set up so cannot re-record those cassettes.

The second I noticed while updating the azure tests for the new fixtures - even after doing that, the tests weren't all passing, and I realized they were also already broken, because F40 EOL recently passed. The stable_nightly_compose fixture used in several azure tests is for F40, and the azure code skips uploading if the EOL date for the release is near or has passed, so now we were never uploading any images for that compose and the tests broke. To fix that we have to update that fixture to be for F42, update the test expectations, and re-record the cassettes. While doing that I noticed the EOL synthesis test was completely broken, so I fixed it too.

With this all tests outside of GCP pass for me locally.