Restructure tests for more intuitive structure

[c-salomonsen]

I did a few things:

1. Removed everyone's individual tests as discussed in Thursdays meeting.
2. Moved tests for wrappers from test_wrappers.py to each corresponding test_model/metric/dataloader.py. Note that this alone added 100+ tests (but not to worry, they run super quick)
3. Rewrote the test logic for our metrics to get better test coverage of the MetricWrapper.

What remains to do:

• Add some validation for each of the metrics, which you, @Seilmast seem to have done in Remaining failing tests #82, could you afterwards perhaps add that to the metricwrapper test?
• Fix the remaining 4 tests that are failing:
  • See below for the new test summary.

=============================================================================== short test summary info ================================================================================
FAILED tests/test_dataloaders.py::test_load_data[svhn-SVHNDataset] - OSError: could not read bytes
FAILED tests/test_models.py::test_load_model[3-image_shape1-magnusmodel] - RuntimeError: mat1 and mat2 shapes cannot be multiplied (5x2352 and 784x133)
FAILED tests/test_models.py::test_load_model[6-image_shape1-magnusmodel] - RuntimeError: mat1 and mat2 shapes cannot be multiplied (5x2352 and 784x133)
FAILED tests/test_models.py::test_load_model[10-image_shape1-magnusmodel] - RuntimeError: mat1 and mat2 shapes cannot be multiplied (5x2352 and 784x133)
4 failed, 160 passed, 48 warnings in 2.24s

[Seilmast]

Seems good!

[Seilmast]

I'll add the missing tests.