Feat/rails graceful shutdown pdb

[binaryknot]

What

Adds three opt-in, values-gated fields to the lago-rails chart (aliased as api and all workers) to harden pods against involuntary disruption during Karpenter node consolidation and EKS upgrades:

• terminationGracePeriodSeconds (pod spec)
• lifecycle (container) — for a preStop sleep
• podDisruptionBudget — new pdb.yaml template with minAvailable / maxUnavailable

Why

During node drains, SIGTERM to Puma and EndpointSlice removal race, so the gateway keeps routing to a terminating pod → brief 502/503s. A preStop sleep keeps the pod serving until it's removed from the gateway's endpoints, and a PDB caps how many pods drain at once. This is the first iteration of hardening we'll roll to staging-eu-1 then prod-eu-1.

Notes for reviewers

• Fully backward-compatible — all three default off/empty, so existing consumers render identically.
• PDB single-replica trap: a PDB on a 1-replica Deployment wedges node drain. Enablement is per-alias via values; we'll only turn it on for the multi-replica api (done in the lago-deploy PR), never blindly on workers.
• Chart.yaml/Chart.lock intentionally untouched — the release pipeline owns version bumps.
• Enabling these in staging-eu-1 / prod-eu-1 (plus the Envoy BackendTrafficPolicy retries) is a separate lago-deploy PR.

Testing

• helm lint clean; 16/16 helm-unittest cases pass (new graceful_shutdown_pdb_test.yaml covers all three, both on and off).
• Verified rendered output for lago-rails and the umbrella lago chart.

[IxDay]

That's literally perfect