Alignment to Original Source datasets for Reproducibility

As noted in #4, all the data paths are hard-coded to OSC (in both CSVs and notebooks) and they don't seem to trace back to files from the existing datasets. For instance, see #4's discussion of `CanonBeetles.csv`. It seems from [this CSV](https://github.com/Imageomics/BeetleVerse/blob/main/Benchmarking/Beetle-PUUM/BeIT-genus.csv) that the Hawaii beetles dataset root should be contained in @SMRayeed's `/fs/ess/PAS2136/Hawaii-2025/beetles_intake/BeetlePUUM/CANON/` folder, but the naming doesn't match the [dataset](https://huggingface.co/datasets/imageomics/Hawaii-beetles/tree/main/individual_specimens). The 2018-NEON Beetles has the same problem. For both, the `individual_images/` directory would seem to correspond to `individual_specimens/`; however, the image names within these folders also do not align.

For the Insect-1M data, it's unclear the intended organization since the `JSON` for their dataset has URLs with paths that don't match how this dataset is structured (based on the included CSVs), and I don't see filenames that match unless they came from somewhere else. I haven't downloaded the [NHM tar (all their images)](https://zenodo.org/records/3549369) to check the structure there, but it seems they have folders based on the GBIF ID, so that is hopefully matchable, but it is currently unclear how one would do that.