fix: add resource limits to all monitoring and ingress components

kube-prometheus-stack deploys operator, alertmanager, kube-state-metrics
and node-exporter without limits by default, causing CPU contention on
the node. Set requests/limits for every component.

NGINX Ingress Controller also lacked resource constraints; add via
Helm --set flags in the deploy job.

Total CPU requests across all components: ~560m / 1930m allocatable
on e2-medium, leaving headroom for kube-system and traffic spikes.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>