jenkins: probe pod CPU allocation before tests

Binding-off did not change per-test times, so the slowdown is the pod's CPU
slice, not core-stacking. Print nproc, cgroup cpu.max/quota, and cpuset before
ctest to find the real allocation, then size -j and ranks to it.

Author: the-hampel

jenkins/Jenkinsfile

@@ -96,6 +96,15 @@ def runBuild = { String platform, String prefix, List<String> envs ->
       """
       catchError(buildResult: 'UNSTABLE', stageResult: 'UNSTABLE') {
         try {
+          /* DIAGNOSTIC: how many CPUs does this build pod actually get? The
+             node is 96-core but the pod gets a cgroup slice; size test
+             parallelism to that, not the host. */
+          sh '''#!/bin/sh -x
+            echo "nproc=$(nproc)"
+            getconf _NPROCESSORS_ONLN || true
+            cat /sys/fs/cgroup/cpu.max 2>/dev/null || cat /sys/fs/cgroup/cpu/cpu.cfs_quota_us /sys/fs/cgroup/cpu/cpu.cfs_period_us 2>/dev/null || true
+            cat /sys/fs/cgroup/cpuset.cpus.effective 2>/dev/null || cat /sys/fs/cgroup/cpuset/cpuset.cpus 2>/dev/null || true
+          '''
           /* Tests run 4 MPI ranks each (PROCESSORS property) with binding
              disabled (see envs), so ctest -j is a core budget: -j16 = 4
              concurrent on Linux, -j8 = 2 on the smaller macOS m1 nodes. */