Test suite to cover typical scenarios on supported datasets

[nmanovic]

To improve testing of the library we need to have a table with experiments for supported datasets.

Experiments:

• Convert from the public dataset to Datumaro format and back to the original format. Metrics: status and correctness of the conversion procedure, time to download, read, write, convert the dataset.
• Take a public dataset and merge it with itself in different modes. Modes: remove duplicates, keep all annotations, merge similar annotations into one. Metrics: status and correctness of the operation, time to merge.
• Extract a subset of a supported dataset, modify the subset, merge it back. Metrics: status and correctness of the operation, time to extract the subset, merge.
• Take a public dataset, modify it, navigate by its history. Metrics: status and correctness of the operation, time to move backward and forward by the history, estimate disk space which is required.
• etc

It will be grade to run the test suite once a week and public results every release. There several benefits here:

• Stress testing on real dataset and scenarios
• Performance testing
• Improve stability and correctness of the library