#4574: - Fix Docker arm64 image crash (`exec format error`) caused by tag race condition

Fixes #4566 ## Problem Both `build-amd64` and `build-arm64` jobs were tagging their images with *all* architecture suffixes. This caused a race condition where whichever job finished last would overwrite the other's tags—so `main-arm64` ended up pointing to an amd64 image. ## Solution Each architecture build job now only tags with its own suffix: - `build-amd64` → `main-amd64`, `v*-amd64` - `build-arm64` → `main-arm64`, `v*-arm64` The `create-manifest` job continues to create the multi-arch manifest (`main`, `v*`) by combining both digests. ## Test plan - [ ] Verify workflow runs successfully on push to main - [ ] Confirm `main-arm64` image runs on arm64 host without exec format error - [ ] Confirm `main-amd64` image runs on amd64 host - [ ] Confirm `main` manifest auto-selects correct architecture <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates `.github/workflows/docker-release.yml` to prevent cross-architecture tag overwrites when building/pushing Docker images to GHCR. The amd64 and arm64 build jobs now only publish architecture-suffixed tags (e.g., `main-amd64`, `vX.Y.Z-amd64` and `main-arm64`, `vX.Y.Z-arm64`), while the `create-manifest` job continues to publish the unsuffixed tags (`main`, `vX.Y.Z`) as a multi-arch manifest composed from the two build digests. This fits the existing release workflow by keeping per-arch images available under explicit tags, and reserving the generic tags for the final manifest list so consumers pulling `:main` or `:v*` get the correct architecture automatically. <h3>Confidence Score: 4/5</h3> - This PR is safe to merge and directly addresses the cross-arch tag overwrite issue with minimal workflow changes. - Changes are narrowly scoped to tag generation in the release workflow and align with the intended model: arch-specific tags per build, generic tags only on the manifest list. The main residual risk is operational (manifest job failure would leave only suffixed tags), not a functional regression in the workflow itself. - No files require special attention <!-- greptile_other_comments_section --> _{(3/5) Reply to the agent's comments like "Can you suggest a fix for this @greptileai?" or ask follow-up questions!} **Context used:** - Context from `dashboard` - CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=fd949e91-5c3a-4ab5-90a1-cbe184fd6ce8)) - Context from `dashboard` - AGENTS.md ([source](https://app.greptile.com/review/custom-context?memory=0d0c8278-ef8e-4d6c-ab21-f5527e322f13)) <!-- /greptile_comment -->