improvement: reduce ARM e2e pod resources to allow concurrent scheduling

[jamesgao-jpg]

Summary

The ARM e2e node (m7gd.2xlarge, 8 vCPU / 32Gi) currently fits only one e2e pod at a time with the current resource requests of 4 CPU / 16Gi. This causes all ARM e2e jobs to queue serially, each taking ~1.5–2h.

Proposal

Reduce ARM e2e pod resource requests from 4 CPU / 16Gi to 3 CPU / 12Gi, allowing two pods to be scheduled concurrently on the same node:

• CPU: 2×3 + 0.6 (system) = 6.6 < 7.9 allocatable
• Memory: 2×12 + 4.5 (system) = 28.5 < 30.5 allocatable

This cuts the serial queue wait time roughly in half.

Related PR

#1569