Parse zip with top level directory

[dtronmans]

Purpose

Make it possible to parse a zip file with a top-level directory instead of immediate sub-directories.

Specification

• Assumes there is only one directory at the root of the zip and nothing else. I filter away __MACOSX, Thumbs.db and desktop.ini as well as anything that could start with . (.DS_STORE, .cache) in case these are created with the top-level dir when the zip file is created.
• The only possible unintended change in parsing is that ClassificationDirectoryParser could falsely win on wrapped archives because it accepts generic class-folder layouts. To avoid that, we unwrap a single top-level folder only when it has explicit dataset-root markers (eg train/valid/test, images/labels, annotations.json, labels.json, data.yaml) or when that inner folder is recognized by any non-CLSDIR parser.
• test_zip_layout_equivalence.py: for different zip datasets, parse the original (flat) zip, parse the zip with an added top-level directory, and parse the extracted zip folder and assert that all 3 return the same LDFs.
• For test_zip_layout_equivalence.py, I used the helper methods from test_parse_export_equivalence.py to assert that the resulting annotations are the same, so since they are used in two tests I moved them to utils.py

Dependencies & Potential Impact

Deployment Plan

Testing & Validation

imagenet-sample zip file with top-level dir, parser detects the correct format:

coco-2017 zip file with top-level dir, parser detects the correct format:

[klemen1999]

Tested on a few examples and LGTM, ty