E2E tests killed by OOM on self-hosted runners

[paolino]

Summary

Linux E2E tests are consistently killed by the kernel OOM killer on self-hosted runners (builder-new). The e2e process is terminated by the memory cgroup limit on the gha-runner service.

Evidence

From dmesg on builder-new:

oom-kill: constraint=CONSTRAINT_MEMCG, cpuset=system-gha\x2drunner.slice
  task_memcg=/system.slice/system-gha\x2drunner.slice/gha-runner@9.service
  task=e2e, pid=766866
Memory cgroup out of memory: Killed process 766866 (e2e)

Options

1. Increase MemoryMax on gha-runner systemd service
2. Split E2E into smaller jobs with lower peak memory
3. Add swap as overflow buffer

Status

• No PR yet
• Not blocking release directly, but blocking CI reliability
• Related to fix: clean up E2E temp directories on all platforms #5196 (temp dir cleanup) and Optimize E2E tests with persistent wallet state per architecture #5155 (persistent wallet state)