By default avoiding generating files in temp directory

[dbogunowicz]

Changes

1. Changed the conftests so that after running any test target, the total size of files created after running the test suite can be: zero mb for the root directory, and up to 1mb for the temp directory.
2. Audited the most important paths for /src and /tests
3. For onnx and python tests reduced the number of data created in the /tmp from 200mb and 800mb respectively down to practically zero mb.

Audit Plan

Auditing the pathways that contain keywords: tempfile and download to make sure that we are not being sloppy with saving or downloading files to the temporary directory.

Auditing the /src

find . -name "*.py" -type f -exec grep -l "download" {} +

./sparseml/onnx/benchmark/info.py - validated
./sparseml/transformers/masked_language_modeling.py - validated
./sparseml/transformers/text_classification.py - validated
./sparseml/transformers/question_answering.py - validated
./sparseml/transformers/token_classification.py - validated
./sparseml/tensorflow_v1/datasets/classification/imagenette.py - validated
./sparseml/tensorflow_v1/datasets/classification/cifar.py - validated
./sparseml/optim/helpers.py - validated
./sparseml/openpifpaf/train.py - validated
./sparseml/pytorch/datasets/detection/coco.py - validated
./sparseml/pytorch/datasets/detection/voc.py - validated
./sparseml/pytorch/datasets/classification/mnist.py - validated
./sparseml/pytorch/datasets/classification/imagenette.py - validated
./sparseml/pytorch/datasets/classification/cifar.py - validated
./sparseml/pytorch/models/external/torchvision.py - validated
./sparseml/pytorch/models/registry.py - validated 
./sparseml/pytorch/torchvision/export_onnx.py - validated
./sparseml/pytorch/torchvision/train.py - validated
./sparseml/pytorch/image_classification/lr_analysis.py - validated
./sparseml/pytorch/image_classification/pr_sensitivity.py - validated
./sparseml/pytorch/image_classification/train.py - validated
./sparseml/pytorch/image_classification/export.py - validated
./sparseml/pytorch/image_classification/utils/helpers.py - validated
./sparseml/pytorch/image_classification/utils/trainer.py - validated
./sparseml/pytorch/utils/helpers.py - validated
./sparseml/pytorch/utils/model.py - validated
./sparseml/benchmark/info.py - validated
./sparseml/yolov8/trainers.py - **validated (allow temp files to be saved (it makes sense))**
./sparseml/keras/datasets/classification/imagenette.py - validated
./sparseml/keras/models/registry.py - validated
./sparseml/utils/datasets/imagenette.py - validated
./sparseml/yolact/scripts.py - validated
./sparseml/yolact/__init__.py - validated

find . -name "*.py" -type f -exec grep -l "tempfile" {} +

./sparseml/onnx/utils/model.py - **validated (I think it should be removed, outdated piece of code))**

Auditing the /tests

Note: I consider saving to tempfile benign if it is either done within a context (using a context manager), or using the tmp_path fixture or factory within the pytest framework.

 find . -name "*.py" -type f -exec grep -l "tempfile" {} +

./integrations/transformers/test_transformers.py - validated 
./integrations/image_classification/test_image_classification.py - validated
./integrations/yolov5/test_yolov5.py - validated
./sparseml/onnx/optim/quantization/helpers.py - validated
./sparseml/onnx/optim/quantization/test_quantize_model_post_training.py - validated
./sparseml/onnx/benchmark/test_info.py - **validated (not fixed)**
./sparseml/onnx/utils/test_data.py - validated
./sparseml/framework/test_info.py - validated
./sparseml/sparsification/test_info.py - validated
./sparseml/optim/test_analyzer.py - validated
./sparseml/optim/test_sensitivity.py - validated
./sparseml/pytorch/sparsification/test_modifier_thinning.py - validated
./sparseml/pytorch/utils/test_helpers.py - validated
./sparseml/pytorch/utils/test_exporter.py - validated
./sparseml/keras/optim/test_mask_pruning.py - validated
./sparseml/yolov5/helpers.py - validated

ubuntu@quad-mle-1:~/damian/sparseml/tests$ find . -name "*.py" -type f -exec grep -l "download" {} +
./integrations/image_classification/args.py - validated
./integrations/image_classification/test_image_classification.py - validated
./sparseml/onnx/benchmark/test_info.py - validated
./sparseml/transformers/utils/test_helpers.py - validated
./sparseml/pytorch/test_torch_to_onnx_exporter.py - validate
./sparseml/pytorch/image_classification/utils/test_helpers.py - validated
./sparseml/yolov5/helpers.py - validated

[bfineran]

LGTM overall - should be noted for testing creating tempfiles is ok

[dbogunowicz]

@bfineran I think that there is a risk (currently low but will grow over time as we work with larger and larger models) of saving very large files to /tmp. This is something that QA already reported. I think that saving files to /tmp is fine, as long as they also get properly cleaned up. We have been a bit sloppy at concluding our unit tests with appropriate cleanups in the past; this PR is a partial remedy to that sloppiness (fun fact, the test that was writing the biggest amount of mbytes to /tmp and did not clean them up was written by the user d.bogunowicz ;))

[KSGulin]

Great cleanup and updated test conditions!