feat(ci): publish binaries for linux/amd64 and linux/arm64 on release

[morgaesis]

Summary

• Add build-binaries job to compile native binaries for Linux x86_64 and ARM64
• Publish binaries as GitHub Release assets when a version tag is pushed
• Use github.repository_owner for dynamic Docker image registry path (fork-friendly)
• Conditionally run Fly registry push and rollout steps only for upstream org

Changes

Binary builds:

• Build for x86_64-unknown-linux-gnu on ubuntu-24.04
• Build for aarch64-unknown-linux-gnu on ubuntu-24.04-arm (native ARM runner)
• Install protoc for Lance dependency
• Package as .tar.gz and upload as release assets

Docker improvements:

• Changed hardcoded ghcr.io/spacedriveapp/spacebot to dynamic ${{ env.REGISTRY }}/${{ github.repository_owner }}/${{ env.IMAGE_NAME }}
• This allows forks to test the full workflow without permission errors
• See also: GitHub Guide - Publishing Docker images which uses ${{ github.repository }} for similar dynamic naming

Conditional deployment:

• Fly registry push and hosted instance rollout only run when github.repository_owner == 'spacedriveapp'

Testing

Tested on fork: https://github.com/morgaesis/spacebot/releases/tag/v0.1.18

[Marenz]

Nice work — the native ARM64 runner is the right call and the fork-friendly registry path is a good touch.

A few suggestions that might help get this merged:

• The version bump commits (0.1.13 → 0.1.19) look like iteration artifacts from testing on your fork — might be worth dropping those and keeping Cargo.toml out of this PR.
• Could be worth squashing the commits down — 17 commits for a ~100 line workflow change makes it a bit hard to review.
• With continue-on-error on the binary builds, a release could be created with missing artifacts silently. Might be worth failing loudly instead so it's obvious when a build didn't produce an artifact.

[morgaesis]

@Marenz
IMO, squashing does nothing, since the PR commit should be a squash commit in itself, not a merge.

That way, whatever dirty work I make on the journey (the PR is the journey), the only commit that matters is the squashed commit on main.

---

Working on the rest, though. Thanks for the feedback ❤️

[morgaesis]

@Marenz, updated according to your suggestions, except for squashing. LMK if you really want this PR squashed.

There is no continue-on-error 🕵️

How do you review code? I find it odd to review individual commits, the thing that matters to me is the diff that enters main, regardless of how we got there. So, when I review, I just check the diff, and maybe the diff since last review; I don't find value in looking at individual commits by others, but maybe that's a mindset I need to change 🤔