Cap pytest-xdist's auto worker count to the container CPU quota

A container's CPU limit is a CFS quota, which nproc and sched_getaffinity do not
see, so --numprocesses=auto sizes itself from the whole machine's core count. On
an 8 vCPU OSDC pod that meant 39+ workers, which exhausted the memory limit and
got the container OOMKilled.

pytest-parallelism.sh reads the quota from cgroup v2 or v1 and exports
PYTEST_XDIST_AUTO_NUM_WORKERS, falling back to nproc where there is no quota, so
it is a no-op off OSDC. Sourced from the two scripts here that run
--numprocesses=auto; backends/arm/test/test_arm_backend.sh follows in the
pull.yml change.

Authored with Claude Code.

Author: huydhn

.ci/scripts/pytest-parallelism.sh

@@ -0,0 +1,37 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# Pin pytest-xdist's `auto` worker count. A container's CPU limit is a CFS quota,
+# which nproc and sched_getaffinity do not see, so `auto` fans out to the whole
+# machine's core count and the workers exhaust the container's memory limit.
+
+pytest_num_workers() {
+  local quota period
+  if [[ -r /sys/fs/cgroup/cpu.max ]]; then  # cgroup v2
+    read -r quota period < /sys/fs/cgroup/cpu.max
+    if [[ "${quota}" != "max" ]] && (( period > 0 )); then
+      echo $(( quota / period ))
+      return
+    fi
+  fi
+  if [[ -r /sys/fs/cgroup/cpu/cpu.cfs_quota_us ]]; then  # cgroup v1
+    quota=$(cat /sys/fs/cgroup/cpu/cpu.cfs_quota_us)
+    period=$(cat /sys/fs/cgroup/cpu/cpu.cfs_period_us)
+    if (( quota > 0 && period > 0 )); then
+      echo $(( quota / period ))
+      return
+    fi
+  fi
+  nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 1
+}
+
+if [[ -z "${PYTEST_XDIST_AUTO_NUM_WORKERS:-}" ]]; then
+  workers=$(pytest_num_workers)
+  (( workers < 1 )) && workers=1
+  export PYTEST_XDIST_AUTO_NUM_WORKERS="${workers}"
+  echo "PYTEST_XDIST_AUTO_NUM_WORKERS=${PYTEST_XDIST_AUTO_NUM_WORKERS}"
+fi

.ci/scripts/test_backend.sh

@@ -7,6 +7,9 @@
 # LICENSE file in the root directory of this source tree.
 set -eux
 
+# Cap pytest-xdist's `auto` workers to the container's CPU quota.
+source .ci/scripts/pytest-parallelism.sh
+
 SUITE=$1
 FLOW=$2
 ARTIFACT_DIR=$3

.ci/scripts/unittest-linux-cmake.sh

@@ -7,6 +7,9 @@
 # LICENSE file in the root directory of this source tree.
 set -eux
 
+# Cap pytest-xdist's `auto` workers to the container's CPU quota.
+source .ci/scripts/pytest-parallelism.sh
+
 # Some ARM/TOSA-adjacent tests import modules that require tosa_serializer.
 # Install from a local tosa-tools checkout when available. If absent in this
 # checkout layout, clone the pinned upstream tag and install from there.