feat: long convergence resiliency for release tests

[ko3n1g]

Summary

This PR improves the resiliency and observability of long-running release test pipelines.

Changes

• ft_launcher for all release tests — adds LAUNCHER: ft_launcher to all 16 _release `model_config.yaml` files so release runs use `ft_launcher` instead of `torchrun`
• Slack alerts during release runs — adds `send_slack_alert()` to `launch_jet_workload.py` that calls `notify.py` at three checkpoints (training finished, pipeline failed/retrying, max attempts exhausted); release-only, all other test types are unaffected
• Golden values — adds/updates golden value files for bert, gpt, mixtral, and t5 release test cases from pipeline 47719739

Affected test cases (ft_launcher)

Model	Test case
BERT	`bert_release`, `bert_release_sm`
GPT	`gpt3_15b_8t_release`, `gpt3_15b_8t_release_gb200`, `gpt3_15b_8t_release_sm`, `gpt3_15b_8t_release_sm_gb200`
Mixtral	`deepseekv3_proxy_flex_tp1pp4emp16etp1cp1_release{,_sm}`, `deepseekv3_proxy_flex_tp2pp2emp16etp1cp1_gb_200_release{,_sm}`, `mixtral_8x22b_tp2pp8ep8vpp1_release`, `mixtral_8x7b_alltoall_tp2pp4ep4_release{,_sm}`, `mixtral_8x7b_tp1pp4ep8vpp8_release`
T5	`t5_release`, `t5_release_sm`

Example diff (ft_launcher)

```yaml
TEST_TYPE: "release"
+LAUNCHER: ft_launcher
MODEL_ARGS:
```

Slack alert flow

```
release run iteration
├── training finished → notify.py (pipeline-context: "<test_case> | iteration=N | attempt=N | training finished")
├── pipeline failed, retry → notify.py (pipeline-context: "<test_case> | iteration=N | attempt=N | pipeline failed, retrying")
└── max attempts exhausted → notify.py (pipeline-context: "<test_case> | iteration=N | attempt=N | max attempts exhausted")
```

Test plan

• CI release pipelines pass with ft_launcher
• Slack alerts fire at the correct checkpoints

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[ko3n1g]

/ok to test

[ko3n1g]

/ok to test

[ko3n1g]

/ok to test

[ko3n1g]

/ok to test

[ko3n1g]

/ok to test

[svcnvidia-nemo-ci]

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24514520678