Add Kaggle test splits

[abidwael]

Adds missing test splits from a number of Kaggle datasets.

[github-actions]

Unit Test Results

         6 files  ±  0           6 suites  ±0   3h 46m 1s ⏱️ + 16m 56s
  3 504 tests ±  0    3 383 ✔️ ±  0    79 💤 ±  0  42 ❌ ±0 
10 456 runs   - 56  10 170 ✔️  - 43  244 💤  - 13  42 ❌ ±0 

For more details on these failures, see this check.

Results for commit 5c27609. ± Comparison against base commit d365d84.

♻️ This comment has been updated with latest results.

[dantreiman]

I think we might want to define a different category for these "test" sets, because these are the test sets for submission to Kaggle and they don't have truth labels included. Thus we can't use them as a "test" split in Ludwig.

Note that train.csv has a loss column (representing the monetary loss from the insurance claim), while test.csv does not

[dantreiman]

I'm concerned that it might mislead users to call splits without a label column "test" -- since in Ludwig's process the test set is a held-out labeled set. We should probably to introduce a new category to support unlabeled data for inference or contest submissions.

[abidwael]

Good catch @dantreiman! I removed the test files from some of the other datasets that had them. We can add a functionality that submits to Kaggle in the future.