Fix save/load profiling notebook and test

[aaron-sandoval]

26 introduced new minimal methods for serializing and deserializing MazeDatasets. However, it was merged with a couple of small outstanding errors:

• The baseline _load_full method was modified, such that no profiling comparison to the legacy loading algorithm was possible. This PR will readd the legacy algorithm and update the profiling notebook to compute improvements based on legacy performance
• The tests for saving and reading datasets only covered the full variants of serializing and deserializing. This PR will add test coverage for the new minimal variants.

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[aaron-sandoval]

5 unit tests not directly related to this PR are mysteriously failing. There appears to be some inconsistencies with results depending on how the tests are run.

• Fail: When run via make test
• Fail: When run via make unit
• Fail: When run in a batch of all tests in test_maze_dataset.py in VSCode
• Pass: When run one by one in VSCode
• Pass: When debugged one by one in VSCode
  • In this case, at least 2 of the tests never reach the line that's failing when run by other methods Failing tests:
• TestMazeDatasetFilters::test_filters
• TestMazeDatasetFilters::test_path_length
• TestMazeDatasetFilters::test_cut_percentile_shortest
• test_maze_dataset.py::test_remove_duplicates
• test_maze_dataset.py::test_remove_duplicates_fast All the tests fail in MazeDataset.collect_generation_meta at line 688:

  >           for key, value in maze.generation_meta.items():
  E           AttributeError: 'NoneType' object has no attribute 'items'

  I suspect there is some subtle thing going on in the test code that's acting statefully and preserving data between tests when it shouldn't be. @mivanit I suppose we could start by seeing if you can replicate the pass/fail behavior above.

[aaron-sandoval]

I ran the tests via make test as suggested since that clears cached data, but the same set of tests fail.