Dataset.from_catalog doesn't use the catalog

[acwooding]

in the docstring is says

If a cached copy of the dataset is present on disk, (and its hashes match those in the dataset catalog), the cached copy will be returned. Otherwise, the dataset will be regenerated by traversing the transformer graph.

However, the code doesn't check for a cached copy. It immediately looks in the transformer graph and then calls a generate which walks the whole graph.