[README] Make basic package description more uniform across different instances; Independent of a particular task

[classify.scripts.saliency] Incorporate evaluation code directly on interpretability/completeness analysis; Remove outdated saliency-evaluation script and engine; Fix test units

[classify.engine.saliency.interpretability] Implement filter for operating on specific datasets only; Implement basic analysis for interpretability

[libs.common.models.transforms] Maximise use of torchvision transforms; Closes #86 as well concluding previous commit