Fix review issues: PREEMPTED state, wait race, concurrency group, retry timeout

- Add PREEMPTED and REVOKED to monitor_slurm_job.sh terminal states
  so preempted jobs don't hang the monitor loop indefinitely
- Wait for both build PIDs unconditionally to prevent orphaned
  processes racing with on_retry_command clean
- Drop event_name from concurrency group so PR and review events
  for the same branch properly cancel each other
- Reduce retry timeout to 150min so retries have room within the
  480min job timeout

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

Author: sbryngelson

.github/scripts/monitor_slurm_job.sh

@@ -58,7 +58,7 @@ get_job_state() {
 # Check if a state is terminal (job is done, for better or worse)
 is_terminal_state() {
   case "$1" in
-    COMPLETED|FAILED|CANCELLED|CANCELLED+|TIMEOUT|OUT_OF_MEMORY|NODE_FAIL|BOOT_FAIL|DEADLINE)
+    COMPLETED|FAILED|CANCELLED|CANCELLED+|TIMEOUT|OUT_OF_MEMORY|NODE_FAIL|BOOT_FAIL|DEADLINE|PREEMPTED|REVOKED)
       return 0 ;;
     *)
       return 1 ;;

.github/workflows/bench.yml

@@ -7,7 +7,7 @@ on:
   workflow_dispatch:
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}
+  group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
 jobs:
@@ -105,13 +105,15 @@ jobs:
         with:
           max_attempts: 3
           retry_wait_seconds: 60
-          timeout_minutes: 480
+          timeout_minutes: 150
           command: |
             (cd pr     && ${{ matrix.build_script }}) &
             pid1=$!
             (cd master && ${{ matrix.build_script }}) &
             pid2=$!
-            wait $pid1 && wait $pid2
+            wait $pid1; e1=$?
+            wait $pid2; e2=$?
+            [ $e1 -eq 0 ] && [ $e2 -eq 0 ]
           on_retry_command: |
             (cd pr     && ./mfc.sh clean) &
             (cd master && ./mfc.sh clean) &