Problem with visual_92_categories dataset

Reported at https://mne.discourse.group/t/mne-visual-92-categories-data-sample-subject-0-raw-1-fif-does-not-exist/5204

The dataset consists of 4 split files, but they've been manually renamed.

To fix:
Each file should be read individually, and on_split_missing='warn' should be passed. Then the file should be re-saved without it getting split. So we're basically manually splitting the files without "linking" them, as is expected by the Representational Similarity Analysis example.

Alternatative:
Actually the decoding for all runs finishes in basically no time on my computer, so there's no need to do this kind of hocus-pocus in the Representational Similarity Analysis example where it allows loading of only a sub-set of runs. Instead, always all runs should be processed. This would then simplify fixing of the dataset: rename the existing files to match the expected split pattern, potentially load & save under a new name, where now MNE ensures that the split naming is done correctly.

Start with the visual_92_categories dataset and the Representational Similarity Analysis example, then inspect how the four split files are named and loaded. Compare the two proposed approaches in the issue and verify that the chosen fix lets the example load the intended data and complete decoding for the required runs.