refactor(ci): Unique env name for images

[andrew-anyscale]

Blocker for spec discovery. wanda files raised in the error message all share the same name attribute of name: "$IMAGE_TO", so spec discovery cannot successfully tie a single name -> Wanda Spec. This property is required, since the discovery uses the name to resolve building pre-req images of form cr.ray.io/rayproject/{name}

Topic: refac-names
Signed-off-by: andrew andrew@anyscale.com

[andrew-anyscale]

Reviews in this chain:
└#60596 refactor(ci): Unique env name for images
　└#60563 feat(ci): add wandaspecs for local discovery
　　└#60564 feat(build): add container-resource-utils
　　　└#60565 feat(build): add build-wheel.sh script for local wheel generation

[andrew-anyscale]

#	head	base	diff	date	summary
0	742e9a61	23402617	diff	Jan 29 18:25 PM	8 files changed, 32 insertions(+), 32 deletions(-)

[gemini-code-assist]

Code Review

This pull request refactors the CI configuration to use unique environment variable names for Docker image identifiers across different build workflows (core, llm, ml, rllib). Previously, a generic $IMAGE_TO variable was used in multiple wanda.yaml files, causing conflicts during the spec discovery phase. By introducing prefixed variables like $CORE_IMAGE_TO, $LLM_IMAGE_TO, etc., each Wanda spec now has a unique name, resolving the issue. The changes are applied consistently across all relevant Buildkite pipeline configurations and Wanda spec files. The refactoring is clear, correct, and directly addresses the problem described.

[aslonnie]

I do not really like this restriction that env vars have to be unique repo wise now.. these env vars should be locally.

we should have a way to separate wanda's input and global variable?

or why do we need this rename? this feels that some abstraction is not doing right.

[andrew-anyscale]

It's only the name variable that has to be globally unique, and it's for spec discovery. This is an issue for these specific cases that define name as the same environment variable IMAGE_FROM

I'm open to ideas, but as-is, we need to be able to find a way to resolve what qyx.wanda.yaml would depend on in the case that we've exported MYENV=foo:

foo.wanda.yaml

name: "foo"

foo-other.wanda.yaml

name: "$MYENV"

qux.wanda.yaml

name: "qux"
froms: cr.ray.io/rayproject/foo

Exercise:

$ MYENV=foo wanda qux.wanda.yaml
# Does this resolve to depending on foo, or foo-other?

---

I'll need ot reflect more on how to not overcomplicate this, but this could maybe be where we depend on some form of params that restricts the Wanda file's name schema