Try garbage collecting to avoid GPU OOM when running the tests

Suggested here:

pytest-dev/pytest#10296 (comment)

seems that explicitly releasing the GPU resources might be necessary to let torch know there is unused memory

stanza/tests/pipeline/test_pipeline_sentiment_processor.py

@@ -1,3 +1,4 @@
+import gc
 
 import pytest
 import stanza
@@ -21,6 +22,7 @@ def pipeline(self):
         """
         A reusable pipeline with the NER module
         """
+        gc.collect()
         return stanza.Pipeline(dir=TEST_MODELS_DIR, processors="tokenize,sentiment")
 
     def test_simple(self, pipeline):