CI: switch to coverage-run pattern to dodge torch+pytest-cov segfault

The hosted Linux runner's torch + Python 3.12.13 + pytest-cov
combination segfaults at conftest-import time when the coverage
tracer starts before torch's C extension finishes initializing.
PR-N1's first 2 CI runs both failed with:

    ImportError while loading conftest 'tests/conftest.py'.
    tests/conftest.py:14: in <module>
        import torch
    E   ValueError: module functions cannot set METH_CLASS or
                   METH_STATIC
    Segmentation fault (core dumped)

The Mac M4 reviewer scripts already use 'coverage run -m pytest'
+ post-hoc 'coverage report --include' to avoid this race
(commit 9d1a250 documented the workaround). Same pattern applied
to the CI workflow here.

Behavior preserved: 100% coverage gate on the same set of modules,
junit.xml + coverage.xml artifacts as before.

Co-authored-by: FluffyAIcode <FluffyAIcode@users.noreply.github.com>

Author: cursoragent

.github/workflows/ci.yaml

@@ -80,7 +80,15 @@ jobs:
           # ``inference_engine.session.coordinator`` and
           # ``inference_engine.session.generator`` — move to the
           # tests/integration/ suite, gated on Mac M4 / CUDA hosts.
-          pytest \
+          #
+          # Coverage is invoked via ``coverage run -m pytest`` rather
+          # than ``pytest --cov=`` to avoid a torch+pytest-cov race
+          # at conftest-import time on the hosted Linux runner
+          # (PR-N1 hit "module functions cannot set METH_CLASS or
+          # METH_STATIC" / SIGSEGV when --cov= started the tracer
+          # before torch's C extension finished initializing).
+          # Same pattern Mac M4 reviewer scripts use.
+          coverage run -m pytest \
             tests/inference_engine/server/ \
             tests/inference_engine/memory/ \
             tests/inference_engine/scheduler/ \
@@ -89,18 +97,13 @@ jobs:
             tests/sdk/python/ \
             tests/training/repr_align/ \
             tests/backends/mlx/test_env.py \
-            --cov=inference_engine.server \
-            --cov=inference_engine.memory \
-            --cov=inference_engine.scheduler \
-            --cov=inference_engine.pipeline \
-            --cov=inference_engine.session.store \
-            --cov=kakeya \
-            --cov=training.repr_align \
-            --cov-report=term \
-            --cov-report=xml:coverage.xml \
-            --cov-fail-under=100 \
             --junitxml=junit.xml \
             -v
+          coverage report \
+            --include='inference_engine/server/*,inference_engine/memory/*,inference_engine/scheduler/*,inference_engine/pipeline/*,inference_engine/session/store.py,sdks/python/kakeya/*,training/repr_align/*' \
+            --fail-under=100
+          coverage xml -o coverage.xml \
+            --include='inference_engine/server/*,inference_engine/memory/*,inference_engine/scheduler/*,inference_engine/pipeline/*,inference_engine/session/store.py,sdks/python/kakeya/*,training/repr_align/*'
 
       - name: Upload coverage artifact
         if: always()