Dataset-related bug fixes, verbosity improvements
This PR contains a variety of fixes related to Dataset as well as some verbosity-related improvements:
- If Dataset is configured multiple times, the previous behavior was to raise an error which seemed sensible. The problem is that this blocks multiple classifiers from being used at one time. Now, reconfiguring a dataset will reset the data transformations allowing to be regenerated when requested.
- Only combine non-empty Datasets within reports. In some edge cases, a Dataset could have no labels in the dataset and the column dtypes were different than for a filled dataset which would cause table merge errors.
- Fix an issue pulling times from Dataset in corner plots.
- Allow Dataset.load_data() to be called with verbose, which propagates to DataLoader.query() so you can see progress as you're loading data.
- Fix an unrelated error in
idq-report
when errors are encountered. The logger wasn't instantiated properly causing a cascading error.