ci: fix stale comments from incremental refactoring

- select-gpu-partition.sh: priority comment said 'smaller/older' but
  list is now L40S/H200/H100/A100/V100
- submit-slurm-job.sh: 'Idempotent' → 'Rerun-safe' (it always
  submits a new job)
- bench.sh: n_jobs only used for build, not bench

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sbryngelson

.github/scripts/select-gpu-partition.sh

@@ -2,8 +2,9 @@
 # Select the best available Phoenix GPU partition using sinfo.
 # Sources into caller: exports SELECTED_GPU_PARTITION.
 #
-# Priority order prefers smaller/older nodes to leave modern GPUs free
-# for production workloads. Falls back to gpu-l40s if nothing is idle.
+# Priority order prefers partitions most likely to have availability.
+# V100 is last due to slower performance near the test time limit.
+# Falls back to gpu-l40s if no partition meets the idle node threshold.
 # RTX 6000 nodes are excluded (too slow for the test suite time limit).
 #
 # Optional: set GPU_PARTITION_MIN_NODES before sourcing to require a minimum

.github/scripts/submit-slurm-job.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 # Unified SLURM job submission and monitoring for all clusters.
 # Submits a script as a SLURM batch job, then monitors it until completion.
-# Idempotent: cancels stale jobs from previous runs before resubmission.
+# Rerun-safe: cancels stale jobs from previous runs before resubmission.
 #
 # Usage: submit-slurm-job.sh <script.sh> <cpu|gpu> <none|acc|omp> <cluster> [shard]
 

.github/workflows/common/bench.sh

@@ -8,7 +8,7 @@ set -euo pipefail
 source .github/scripts/bench-preamble.sh
 
 # Cap parallel jobs at 64 to avoid overwhelming MPI daemons on large nodes
-# (GNR nodes have 192 cores but nproc is too aggressive for build/bench).
+# (GNR nodes have 192 cores but nproc is too aggressive for build).
 n_jobs=$(( $(nproc) > 64 ? 64 : $(nproc) ))
 
 # --- Phoenix TMPDIR setup ---