Adds opt-in, lease-based leader election so running multiple replicas no longer multiplies load on the cloud provider pricing APIs. Off by default, so single-replica deployments are unchanged.

When enabled (-leader-election.enabled):

• Replicas elect one leader via a Kubernetes Lease.
• The leader registers the provider collectors and calls the cloud provider APIs.
• Non-leaders serve /metrics with the base operational metrics plus a cloudcost_exporter_leader_election_is_leader gauge, but do not call the provider APIs.
• Losing the lease stops collection and shuts the replica down so it rejoins as a candidate.

The result is a single set of upstream API calls regardless of replica count. Sharding (splitting work across replicas) is intentionally out of scope, per the issue.

Validation

• make lint — 0 issues
• go test ./... — all pass (new pkg/leaderelection tests including a fake-clientset leadership-acquisition test; updated cmd/exporter tests for the split registry handler)
• make build-binary — success
• New dependencies: k8s.io/client-go, k8s.io/apimachinery, k8s.io/api @ v0.34.2 (stable)

Docs

• docs/deploying/leader-election.md — flags, required RBAC (leases in coordination.k8s.io), and the leader gauge
• README link from the Kubernetes deployment section

closes: #1027