Classifier gives inconsistent classifications

[kirstybayliss]

The classifier sometimes gives inconsistent event classifications. The profile_workflow_classification test highlighted this, passing locally and in the github CI checks but failing elsewhere. For Chris (on windows), there are 12 events (in ~4600) that switch between 'slab' and 'unknown':

Reducing the test data to exclude these mismatching events still results in a failure for Michele (but not locally for me or Chris, or in the CI).
Michele has suggested this might be due to floating point rounding, which is often machine dependent. Either way, we should add more tests to the classification step. In the meantime, the profile_workflow_classification test is skipped.

The original classification (for me) and Chris's classifications are in the csv, it might be useful to compare with any other classification outputs people get.

classified_earthquakes_test_orig.csv