[DYNAMO] add SLURM smoke config for Dynamo

[AmeenP]

Summary

Adds a single-node SLURM smoke example for running prime-rl against NVIDIA Dynamo as the inference backend.

This is stacked on #2394 so it lives next to the existing Dynamo K8s/local smoke examples.

Files added:

• tools/dynamo/slurm_smoke/smoke_rl.toml: RL config for a 2-GPU smoke run with no prime-rl-managed inference process.
• tools/dynamo/slurm_smoke/dynamo_single_node_rl.sbatch.j2: custom sbatch template that starts Dynamo on GPU 0 with --discovery-backend file, waits for /health and /v1/models, then runs prime-rl on GPU 1.
• tools/dynamo/slurm_smoke/README.md: dry-run, submit, environment override, and log instructions.

The Dynamo side uses file discovery and disables KV event publishing, so the sample does not require etcd or NATS for the single-node smoke path.

Validation

Local:

• Parsed tools/dynamo/slurm_smoke/smoke_rl.toml through RLConfig using the config package and temporary validation dependencies.
• Rendered dynamo_single_node_rl.sbatch.j2 with Jinja2 and checked the rendered script with bash -n.
• git diff --cached --check

GitHub Actions:

• Ruff: pass
• Slim install (prime-rl-configs only): pass
• Unit tests: pass

Note: full uv run rl ... --dry-run cannot run in this macOS workspace because prime-rl's lockfile supports Linux platforms only, and the local environment does not have the project console scripts synced.