Investigate warnings when splitting MMLV data

[schrum2]

MarioDiffusion>python split_data.py --json_file Game_MMLV\DATA\MMLV_LevelsAndCaptions-regular.json --train_pct 0.9 --val_pct 0.05 --test_pct 0.05 --seed 42 --game MMLV
Train set saved to: Game_MMLV\DATA\MMLV_LevelsAndCaptions-regular-train.json (65147 samples)
Validation set saved to: Game_MMLV\DATA\MMLV_LevelsAndCaptions-regular-validate.json (3619 samples)
Test set saved to: Game_MMLV\DATA\MMLV_LevelsAndCaptions-regular-test.json (3620 samples)
WARNING: required structure 'powerup' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'enem' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'hazard' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'disappearing block' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'water' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'moving platform' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'platform' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'ladder' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'loose block' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'floor' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'ceiling' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'wall' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'rectangular block cluster' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'irregular block cluster' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.
WARNING: required structure 'tower' is absent from the dataset; it cannot be placed in all 3 splits, so its coverage is not enforced.

Based on MMLV data from following instructions in Game_MMLV/README.md

[schrum2]

I think the cause of these warnings is that this particular dataset was made with the captions in the "deterministic_captions" key rather than the usual "caption" key, so it probably can't find any captions. We should add a way around this that still checks that the split is good.

In contrast, we'll probably just have to completely ignore these checks when dealing with LLM generated captions, but since we know that, we should make it so the warnings don't show up in that particular case.

[schrum2]

Basically, the warnings don't matter when using caption keys, so make the code not give the warnings in this case.