Add wanda anyscale image builds for release tests

[andrew-anyscale]

This adds wanda-based builds for anyscale images used in release tests.

Changes:

• Add ray-anyscale-cpu/cuda.wanda.yaml for ray anyscale images
• Add ray-llm-anyscale-cuda.wanda.yaml for LLM images
• Add ray-ml-anyscale-cuda.wanda.yaml for ML images
• Add push_anyscale_image.py for pushing to ECR/GCP/Azure via crane
• Update release/build.rayci.yml with build and push steps
• Update gcloud_docker_login.sh to v550.0.0 with arch detection

Topic: anyscale-image
Relative: ray-wheel
Labels: draft

[andrew-anyscale]

Reviews in this chain:
└#59935 Migrate wheel builds to wanda (x86_64)
　└#59969 Add wanda cpp wheel builds (x86_64)
　　└#59936 Add wanda ray image builds for Docker Hub
　　　└#59937 Add wanda anyscale image builds for release tests

[andrew-anyscale]

#	head	base	diff	date	summary
0	f963b425	f104bf59	diff	Jan 7 7:57 AM	8 files changed, 553 insertions(+), 24 deletions(-)
1	07109714	288cadb6	diff	Jan 7 7:58 AM	0 files changed
2	4cdb2dd9	46cc8625	diff	Jan 7 8:38 AM	2 files changed, 238 insertions(+), 5 deletions(-)
3	f65669a5	70df7561	rebase	Jan 7 11:29 AM	0 files changed
4	11418977	ae32d45a	rebase	Jan 7 12:21 PM	0 files changed
5	af3b600e	04cfd26a	rebase	Jan 7 17:29 PM	0 files changed
6	6bbeb9e4	55d1473f	rebase	Jan 7 17:31 PM	0 files changed
7	1f049e81	84d74797	diff	Jan 8 8:27 AM	2 files changed, 12 insertions(+), 12 deletions(-)
8	d43ccd7f	c275fe24	rebase	Jan 8 8:32 AM	0 files changed
9	8980d9ce	7c78fa2d	rebase	Jan 8 12:01 PM	0 files changed
10	4a143b7a	29cf5abd	rebase	Jan 8 12:51 PM	0 files changed
11	9243672a	d566c8a8	rebase	Jan 8 13:16 PM	0 files changed
12	8c684c8e	91900297	rebase	Jan 8 13:33 PM	0 files changed
13	826bac0f	73822f53	rebase	Jan 8 14:30 PM	0 files changed
14	ebdca80e	94805e95	rebase	Jan 9 8:45 AM	0 files changed
15	35249bf7	fc79d67d	rebase	Jan 9 10:01 AM	0 files changed
16	0e1bd8b1	db94f935	rebase	Jan 9 10:59 AM	0 files changed
17	5de31a90	05c06707	rebase	Jan 9 12:36 PM	0 files changed
18	a96c770e	a6718508	rebase	Jan 9 13:32 PM	0 files changed

[gemini-code-assist]

Code Review

This pull request refactors the Anyscale image build process to use Wanda for building and a new Python script for pushing images to multiple registries. The changes include new Wanda configuration files, the push_anyscale_image.py script, and updates to the Buildkite pipeline. My review identifies a few critical issues that will break the build, including a missing file in a Bazel target and an incomplete matrix configuration in the pipeline. I've also provided several suggestions to improve maintainability by reducing code duplication and making parts of the new script less brittle.

[gemini-code-assist]

This py_binary target for push_ray_image references a source file push_ray_image.py that is not included in this pull request. This will cause the Bazel build to fail. Please either add the missing file or remove this target if it's not intended to be part of this change.

[gemini-code-assist]

These authentication commands are duplicated across the anyscalebuild, anyscalellmbuild, and anyscalemlbuild push jobs. To improve maintainability and reduce redundancy, consider using YAML anchors and aliases to define these commands once and reuse them in each job. This will make future updates to the authentication process much easier.

[gemini-code-assist]

The GPU_PLATFORM is hardcoded here. The comment mentions it must match a definition in another file, which makes this brittle. If the default GPU platform changes, this script will require a manual update, which can be easily missed. Consider sourcing this value from a shared configuration file or passing it as a command-line argument to make the script more robust and maintainable.

[gemini-code-assist]

The logic for parsing the CUDA version string using platform.split('.') is fragile. It assumes a specific format with dots as separators and might fail with an IndexError if the platform string format changes (e.g., cu12-cudnn8). Using a regular expression would be more robust for extracting the version parts. For example: re.match(r"cu(\\d+)\\.(\\d+)", platform).

[gemini-code-assist]

This function has a lot of repeated code, especially the block for appending rayci_build_id tags, which is identical in all if/elif/else branches. This can be refactored to be more DRY (Don't Repeat Yourself), which improves readability and maintainability. You could determine the tag prefix first, then generate the tags in a single block of code.

def _get_image_tags(python_version: str, platform: str) -> List[str]:
    """
    Generate image tags matching the original docker_container.py format.

    Returns multiple tags for the image (canonical + aliases).
    For GPU_PLATFORM, also generates -gpu alias tags to match release test expectations.
    """
    branch = os.environ.get("BUILDKITE_BRANCH", "")
    commit = os.environ.get("BUILDKITE_COMMIT", "")[:6]
    rayci_build_id = os.environ.get("RAYCI_BUILD_ID", "")

    py_tag = _format_python_version_tag(python_version)
    platform_tag = _format_platform_tag(platform)

    # For GPU_PLATFORM, also create -gpu alias (release tests use type: gpu)
    platform_tags = [platform_tag]
    if platform == GPU_PLATFORM:
        platform_tags.append("-gpu")

    tags = []
    if branch == "master":
        prefix = commit
    elif branch.startswith("releases/"):
        release_name = branch[len("releases/") :]
        prefix = f"{release_name}.{commit}"
    else:
        pr = os.environ.get("BUILDKITE_PULL_REQUEST", "false")
        if pr != "false":
            prefix = f"pr-{pr}.{commit}"
        else:
            prefix = commit

    for ptag in platform_tags:
        tags.append(f"{prefix}{py_tag}{ptag}")

    if rayci_build_id:
        for ptag in platform_tags:
            tags.append(f"{rayci_build_id}{py_tag}{ptag}")

    return tags