implement KAI-Scheduler backend

[daisy-ycguo]

What type of PR is this?

This PR implements the KAI scheduler backend SyncPodGang() flow in Grove’s scheduler backend framework.
Specifically, it:

• Adds PodGang -> KAI PodGroup translation and reconciliation (create/update) in operator/internal/scheduler/kai/backend.go
• Adds PodGroup cleanup on PodGang deletion via OnPodGangDelete()
• Adds migration-safe behavior by annotating PodGang with grove.io/ignore: "true" so legacy KAI Grove podgrouper can ignore these PodGangs
• Registers KAI PodGroup API types in Grove scheme (scheduling.run.ai/v2alpha2)
• Adds RBAC permissions for scheduling.run.ai/podgroups in the Helm ClusterRole
• Adds/extends unit tests for KAI backend sync and delete paths

What this PR does / why we need it:

• Grove now owns backend-specific scheduling resources directly for KAI.
• Without this, KAI backend remains incomplete and relies on legacy plugin behavior.
• The ignore annotation prevents duplicate/competing PodGroup management during migration.

Which issue(s) this PR fixes:

Fixes #525

Special notes for your reviewer:

• This change is designed to work with KAI-Scheduler support for ignoring PodGangs annotated with grove.io/ignore:
  • Allow Grove plugin to ignore PodGangs with specific annotation kai-scheduler/KAI-Scheduler#1001
• The new KAI backend path reuses the same topology/subgroup semantics used by the legacy Grove plugin mapping, but moves ownership into Grove operator backend reconciliation.

Does this PR introduce a API change?

NONE

Additional documentation e.g., enhancement proposals, usage docs, etc.:

NONE

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[enoodle]

Hi @daisy-ycguo
I think that today KAI doesn't know to read this new PodGroup and it will still create its own PodGroup and ignore Grove's one.
We should work on updating KAI to allow using an external PodGroup like this one and then it will have more effect.
I opened an issue in KAI for that: kai-scheduler/KAI-Scheduler#1420

[enoodle]

Now I see the updated version of the PR message so I think we are on the same page on that 👍

[kangclzjc]

Hi @daisy-ycguo I think that today KAI doesn't know to read this new PodGroup and it will still create its own PodGroup and ignore Grove's one. We should work on updating KAI to allow using an external PodGroup like this one and then it will have more effect. I opened an issue in KAI for that: kai-scheduler/KAI-Scheduler#1420

@enoodle Sanjay has an PR to fix this, could you please check whether it can help