feat(rescheduling): gpuFragmentation eligible by default + env kill switch

[pfernandes21]

What type of PR is this?

/kind feature

What this PR does / why we need it:

Two changes to the gpuFragmentation repack strategy:

1. Eligibility flips from opt-in to opt-out. Previously a pod was only movable if it carried exa.ai/repack-eligible=true; now every pod is eligible by default, and setting exa.ai/repack-eligible=false excludes it:

-	EligibleLabel   string `mapstructure:"eligibleLabel"`   // require label == "true"
+	OptOutLabel     string `mapstructure:"optOutLabel"`     // skip when label == "false"

-	if sole.Pod.Labels[conf.EligibleLabel] != "true" {
+	if sole.Pod.Labels[conf.OptOutLabel] == "false" {
 		return nil
 	}

2. Env-var kill switch. Setting EXA_GPU_REPACK_DISABLED=true on the scheduler process makes the strategy return no victims, independent of the scheduler configmap:

kubectl -n volcano set env deploy/<scheduler> EXA_GPU_REPACK_DISABLED=true   # off (pod restarts)
kubectl -n volcano set env deploy/<scheduler> EXA_GPU_REPACK_DISABLED-      # back on

All other safety gates are unchanged and still bound the blast radius: only negative-priority pods (maxVictimPriority: -1 default — priority ≥ 0 is never touched), sole GPU pod on its node, controller-owned, single-member PodGroup, no karpenter.sh/do-not-disrupt, strictly fuller feasible destination, pool cooldown, per-PodGroup eviction cap, and one victim per pool per pass.

Which issue(s) this PR fixes:

Fixes #

Special notes for your reviewer:

Behavioral widening: after rollout, any negative-priority sole-GPU pod becomes repackable without labeling. Workloads that must never be repacked should set exa.ai/repack-eligible=false (or a priority class ≥ 0, or karpenter.sh/do-not-disrupt). Unit tests updated: unlabeled pod is now selected; =false is skipped; 14/14 pass.

Does this PR introduce a user-facing change?

gpuFragmentation repacking is now opt-out (exa.ai/repack-eligible=false) instead of opt-in, and can be disabled entirely with EXA_GPU_REPACK_DISABLED=true on the scheduler.

Link to Devin session: https://app.devin.ai/sessions/6351ff07ed074c55a521b26d6af34c1b
Requested by: @pfernandes21

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring