OnnxRuntime tests refactor to make them faster (#315)
* Stop downloading for optimization test, and test on a lot less examples for the modeling tests
* Fuse some tests
* Add gc collect
* Change camembert and roberta model name
* Change the way model architectures are fed to the tests
* Fix test on Windows
* Use multiprocessing for onnxruntime tests
* Update github action step name
* Update github actions to use macos-latest
* Made test_optimization lighter
* Fix multiple workers for pytest
* Make test_pipeline_model_is_none for Seq2Seq faster
* Trigger CI
* Fix test_pipeline_model_is_none
* Change any to all
