Severe slowdown in `scipy.stats.pearsonr` calls when using Dask (via result_caching)

When the `dask` library is installed as a dependency (through the `result_caching` module, see [here](https://github.com/brain-score/result_caching/blob/31ea59215e8a81ed3b6720dbe6eef3e9fd065f6e/setup.py#L12)), calls to `scipy.stats.pearsonr` become significantly slower. This slowdown notably impacts performance during ceiling calculations, which rely on `result_caching`. Removing or bypassing `dask` in this workflow returns `pearsonr` to its expected speed. We need to investigate why `dask` is causing this overhead and how to mitigate it without sacrificing functionality in `result_caching`.