feat: publish main-flavor zfs akmods from this fork

[Danathar]

What

Makes this fork publish the akmods images zfs-aurora-complex needs, instead of that repo rebuilding the ZFS akmod from this repo's source on every run.

• org: ublue-os → org: danathar in the akmods-base anchor. Every image ref is assembled as <registry>/<org>/<name>; left at ublue-os, just push from this fork gets a 403.
• zfs target added to 43.main and 44.main by merging in the existing *server-build-group-only-x86 anchor. Upstream builds zfs only for the server flavors, but Aurora DX rides the Fedora main kernel. Produces ghcr.io/danathar/akmods-zfs:main-43 and :main-44.
• zfs.main.linux_experimental: true. OpenZFS 2.4.3 refuses to configure against a kernel newer than 7.0 and F44's main kernel is 7.1.x:

  configure: error:
      *** Cannot build against kernel version 7.1.4-202.fc44.x86_64.
      *** The maximum supported kernel version is 7.0.
  

  This is the same failure currently breaking zfs-aurora-complex's nightly (runs 30148339311, 30192096209). Same override coreos-testing already carries.
• build-akmods-main.yml regenerated with just generate-workflows. The generator randomizes the cron minute on every run, so the schedule was restored to its existing value and the four unrelated flavor workflows reverted — the only real change is the two new build-main_*_zfs jobs and their addition to the check-main_* gates.
• FORK-PATCHES.md added: a registry of every deliberate delta against ublue-os/akmods, split into permanent fork identity vs. temporary workarounds, each temporary one with a stated removal condition.

Caveats

• --enable-linux-experimental disables an upstream compatibility gate. The module may build and still misbehave against a kernel OpenZFS has not validated. Documented as T2 in FORK-PATCHES.md with a removal condition.
• No repository secrets are configured, so these builds fall back to the test signing key (certs/*.priv.test) and pushed images are not cosign-signed. Noted in FORK-PATCHES.md.

Testing

Build MAIN akmods was disabled_manually (since the 2026-07-05 scheduled runs hit a transient kojipkgs truncation) and has been re-enabled, so this PR exercises the full matrix. PR runs build and test but skip the push, so this validates the build; publishing requires a workflow_dispatch on main after merge.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b68b856e15

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Remove the nonexistent fork README patch

The registry instructs future upstream merges to preserve a fork banner and ghcr.io/danathar/... examples, but the current README.md still has the # ublue-os akmods heading and ghcr.io/ublue-os/... examples. Because this file is intended to be the authoritative checklist for mechanical upstream merges, this false entry can cause maintainers to discard upstream README changes while trying to preserve content that does not exist; either apply the described README customization or replace this entry with the actual README delta.

Useful? React with 👍 / 👎.