feat: add automatic deploy dependency detection for KEDA ScaledObjects

[Sharvash]

nelm now automatically deploys a KEDA ScaledObject after the workload referenced by its spec.scaleTargetRef, so the ordering no longer has to be declared by hand.

Previously, when a ScaledObject and its target workload (e.g. a Deployment) belonged to the same release, nelm put both in the same deploy stage with no ordering between them. If the ScaledObject reached the API server before its target existed, the install broke in one of two ways:

• with a cpu/memory trigger, KEDA's validating webhook resolves the target and denies the request (the error reported in KEDA ScaledObject is trying to be deployed before deployment #456): admission webhook "vscaledobject.kb.io" denied the request: Deployment.apps "app" not found
• with any other trigger the object is admitted, but the KEDA controller cannot reconcile it and it stays Ready=False with reason=ScaledObjectCheckFailed

The only way around it was a manual dependency annotation:

werf.io/deploy-dependency-app: state=present,kind=Deployment,name=app

internalDeployDependencies already derives implicit dependencies for built-in workloads, RBAC bindings and others. This adds a keda.sh/ScaledObject case that reads spec.scaleTargetRef and emits a present dependency on the referenced workload:

• name - required; if it is missing the ScaledObject is left untouched
• kind - defaults to Deployment
• apiVersion - defaults to apps/v1, and the group is parsed from it

The dependency is present, not ready: nelm waits only for the target to exist, which is all KEDA needs, instead of blocking on its readiness. ScaledJob is intentionally left out — it uses jobTargetRef and creates its own Jobs, so it has no such dependency.

Verified against a live cluster running KEDA: with the change the ScaledObject is always created after its target and reconciles cleanly, and the #456 webhook denial reproduces without it.

Closes #456

[ilya-lesikov]

Hi, please, resolve conflicts, but don't force-push anymore, otherwise I need to look at all the commits again.

[ilya-lesikov]

Thanks!