Arm backend: Add serialized xlarge TOSA model suite (pytorch#21492)

The DeepSeek-R1-Distill-Qwen layer-test PRs were reverted because they
caused OOMs in the no-driver Arm TOSA model pytest job.

The normal TOSA model shard should keep using pytest-xdist auto
parallelism so unrelated model tests do not slow down. Add a separate
serialized suite for xlarge TOSA model tests instead.

The reland of the DeepSeek layer tests can mark the memory-heavy cases
as xlarge and route them through this suite. This also gives internal CI
an explicit option for running any xlarge TOSA model coverage serially.

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils
@Sebastian-Larsson @robell @rascani

Signed-off-by: Baris Demir <baris.demir@arm.com>

Author: bdemirb

backends/arm/test/test_arm_backend.sh

@@ -115,6 +115,18 @@ test_pytest_models_tosa() {
     echo "${TEST_SUITE_NAME}: PASS"
 }
 
+test_pytest_models_tosa_xlarge() {
+    echo "${TEST_SUITE_NAME}: Run xlarge pytest models for TOSA"
+
+    # Install model dependencies for pytest
+    source backends/arm/scripts/install_models_for_test.sh
+
+    pytest "${PYTEST_RETRY_ARGS[@]}" --verbose --color=yes \
+        --numprocesses=1 --durations=0 \
+        backends/arm/test/models -k tosa -m xlarge
+    echo "${TEST_SUITE_NAME}: PASS"
+}
+
 test_run_tosa() {
     echo "${TEST_SUITE_NAME}: Test TOSA delegate examples with run.sh"