chore(helm): configurable lifecycle, terminationGracePeriodSeconds and autoscaling behavior

[kylegato-dutchie]

Adds three pass-through values so the chart can express graceful pod shutdown and HPA scaling behavior:

Value	Renders into	Default
lifecycle	container lifecycle	{}
terminationGracePeriodSeconds	pod terminationGracePeriodSeconds	''
autoscaling.behavior	HPA spec.behavior	{}

All three default to empty, so rendered output is unchanged for existing users.

Motivation

Without a preStop hook there's no way to drain a pod gracefully. That matters more than it might seem, because HPA scale-down deletes pods directly rather than through the eviction API, so a PodDisruptionBudget doesn't apply to it.

On our deployment of this chart behind an ingress controller, that produced HTTP 504s on every scale-down: the container got SIGTERM while the ingress was still routing to it, because Service endpoint removal is asynchronous. Correlation was unambiguous — the HPA oscillated 2→4→2, and 504 bursts landed only in the scale-down windows, never on scale-up, with no OOMKills, no restarts and clean application logs.

A preStop sleep keeps the pod serving until endpoint removal propagates, and autoscaling.behavior lets operators damp the oscillation that triggers a drain in the first place. These were the only knobs missing to fix it from values — everything else needed was already configurable.

Example

terminationGracePeriodSeconds: 60
lifecycle:
  preStop:
    exec:
      command: ["/bin/sh", "-c", "sleep 30"]
autoscaling:
  enabled: true
  behavior:
    scaleDown:
      stabilizationWindowSeconds: 900
      policies:
        - type: Pods
          value: 1
          periodSeconds: 300

terminationGracePeriodSeconds should exceed any preStop sleep, or the kubelet SIGKILLs the container mid-hook. Both notes are in the values comments.

Verification

• helm lint clean.
• Backward compatibility: helm template with default values, diffed against the previous chart — identical apart from the helm.sh/chart version label.
• All three values render into the expected fields.
• Rendered objects accepted by a Kubernetes 1.35 API server via kubectl apply --dry-run=server.

Chart version 1.2.0 → 1.3.0 (additive, no breaking change), following the convention from #258.

Happy to adjust naming or placement — I put lifecycle and terminationGracePeriodSeconds next to podDisruptionBudget since they're all pod-lifecycle concerns, and behavior inside the existing autoscaling block.