Extend dataset export functionality to all different DatasetTypes

[dtronmans]

Purpose

Go from only being able to export with DatasetType.NATIVE to all DatasetType entries.

Specification

• Every current implemented parser now has a matching exporter.
• The logic for preparing the LDF (shared across all exporters), transforming it and saving are all separated
• BaseExporter defined to be implemented by different exporters, in a very similar way that BaseParsers defines an interface for different parsers

Dependencies & Potential Impact

Deployment Plan

Testing & Validation

The first way I tested the correctness of the functionality is to parse datasets, export them using the exporters then open them with different tools like Roboflow and FiftyOne:

The second and more precise way is by writing tests in tests/test_data/test_parse_export_equivalence.py, in which the following is tested:

1. Parse dataset collected from Roboflow, Google Cloud, etc... and save as prev_dataset
2. Export the dataset using defined exporters
3. Parse this exported dataset again and save as new_dataset
4. Assert that previous and new dataset are exactly the same. For bounding boxes, even though I try to save with as much precision as possible (.12f), we sometimes lose precision with repeated parsing and exporting so I assert that the normalized boxes are almost equal by rounding to 2 decimal places.

[kozlov721]

Thamks. Left some few comments. Otherwise I think it looks good