Final SLURM fixes

[allenwang28]

it seems with the Monarch pin update, we need to use

MONARCH_HOST_MESH_V1_REMOVE_ME_BEFORE_RELEASE

rather than

MONARCH_HOSTMESH_V1

Additionally, spawn_procs APIs match now. Tomorrow we can probably just remove the V1 path altogether, but keeping it for now just to make sure things are working.

With this, I see that SLURM 32B works:

MONARCH_HOST_MESH_V1_REMOVE_ME_BEFORE_RELEASE=1 TORCHSTORE_RDMA_ENABLED=1 python -m apps.grpo.main --config=apps/grpo/qwen3_32b.yaml

Note:

  rl_trainer_perf/push_weights/ts_save/duration_avg_s: 9.150277181121055
  rl_trainer_perf/push_weights/ts_save/duration_max_s: 9.425452718045563
  policy_worker_perf/update_weights/total_duration_avg_s: 77.36585016347817
  policy_worker_perf/update_weights/total_duration_max_s: 88.17838381230831

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@4b3b3c2). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #398   +/-   ##
=======================================
  Coverage        ?   64.70%           
=======================================
  Files           ?       79           
  Lines           ?     7700           
  Branches        ?        0           
=======================================
  Hits            ?     4982           
  Misses          ?     2718           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.