Reduce Java service memory limits

[kenahrens]

Summary

• Heap 192→128MB, metaspace 160→128MB across all 3 Java services (api-gateway, accounts, transactions)
• K8s memory limit 768→512Mi per pod, saving ~768Mi total across the 3 services
• Dockerfile MaxRAMPercentage 75→50% as fallback default
• Current usage is ~500Mi per pod at 2 req/s demo traffic — way overprovisioned

Test plan

• Pods start without OOMKill on dev cluster
• Memory usage stabilizes under 400Mi per pod
• No increase in request errors or latency

🤖 Generated with Claude Code