feat(slurm): AUT-1454 persist multiplexed ssh connections

[svcnemo-autobot]

Background

SlurmExecutor's Paramiko transport cannot be reused after its Python process exits. Repeated NeMo Run invocations therefore establish new SSH connections.

What changed

• Add opt-in OpenSSH multiplexing through control_persist and an optional control_path.
• Reuse a stable control master across NeMo Run processes for commands, SCP, rsync, and port forwarding.
• Preserve existing Fabric/Paramiko behavior when multiplexing is not configured.
• Add no dependencies.

Details

flowchart LR
  N1[NeMo Run process] --> M[OpenSSH control master]
  N2[Later NeMo Run process] --> M
  M --> S[Slurm host]

Loading

The master intentionally remains after Python cleanup and exits after the configured OpenSSH ControlPersist idle duration.

Tested

• Whole-tree Ruff format and lint
• 59 focused tunnel/rsync tests
• 1,352 non-Docker tests with coverage on the implementation head
• Tunnel configuration reconstruction smoke test

The Docker-scheduler module requires an unavailable Docker daemon. End-to-end Slurm validation was not performed, per task constraints.