mecak8s: make Kubernetes Lease renewal capacity explicit and scalable

[ChrisJBurns]

Problem

The Kubernetes client uses client-go's default 5 QPS / 10 burst limits. Each held session Lease renewal performs a GET followed by an UPDATE, normally every ten seconds. Roughly 25 concurrently held sessions consume the entire steady-state client budget before acquisitions, releases, scheduler election, retries, or other Kubernetes work.

Throttled renewals can miss the 30-second Lease TTL, cancel live work, and trigger avoidable lease-loss recovery.

Evidence

• internal/app/build.go:newK8sClientset does not set rest.Config.QPS or Burst.
• internal/adapter/k8slease/k8slease.go:Renew performs GET + UPDATE.
• cmd/mecak8s/flags.go defaults the Lease TTL to 30 seconds and renewal interval to TTL / 3.

Proposed direction

Define and document a capacity model. Make QPS/Burst explicit and configurable or derive safe values from supported concurrency. Consider reducing renewal round trips or moving high-cardinality session fencing away from the Kubernetes API.

Acceptance criteria

• Supported concurrent-session capacity is documented.
• Client throttling settings are explicit and observable.
• A load test exercises acquisitions and renewals at the supported limit.
• Sustained supported load does not cause renewal loss or live-run cancellation.