Update Ray's version to support label-based scheduling + autoscaler

[Future-Outlier]

After this commit is released in Ray’s official image, we should update all Ray images in our examples and default configurations to use it.
ray-project/ray#56532

Discussion link: #4106 (review)
release ref: https://github.com/ray-project/ray/releases

[Future-Outlier]

We should also update the doc on how to use ray start params with {Head|Worker}GroupSpec.Labels.

[ryanaoleary]

ray-project/ray#56532 reads the --labels field from the rayStartParams, with #4106 we'll need to instead check the top level Labels field, which will require another change to Ray. I can try to get that in in the next release of Ray after we've added the new .Labels structured field.

[Future-Outlier]

ray-project/ray#56532 reads the --labels field from the rayStartParams, with #4106 we'll need to instead check the top level Labels field, which will require another change to Ray. I can try to get that in in the next release of Ray after we've added the new .Labels structured field.

Thank you Ryan, really appreciate this

[Future-Outlier]

Hi, @ryanaoleary
I also need you create a PR about "[Core][Autoscaler] Add resources to KubeRay autoscaling config" in ray.
now we only support labels in ray.

[ryanaoleary]

Hi, @ryanaoleary I also need you create a PR about "[Core][Autoscaler] Add resources to KubeRay autoscaling config" in ray. now we only support labels in ray.

@Future-Outlier I implemented support for both the new top level fields in ray-project/ray#57260, adding them to the KubeRay autoscaling config if found. There was already support for parsing the resources (here) and labels (here) from the rayStartParams or k8s container, but now we first check the new resources and labels fields respectively since they are the most explicit.