Eliminate DataDirectory and Dataset Registration

[mdr223]

One of the biggest hurdles to users getting started quickly is the need for them to register datasets with PZ.

While dataset registration makes it easier for the system to track the lineage of computation:

1. Caching is nowhere near being fully supported in PZ
2. In an ideal world, PZ can still cache intermediate results effectively without always requiring users to register datasets and provide dataset ids

To emphasize this latter point: a user who wants to quickly test out PZ should not need to provide a dataset_id (or register their dataset) just because some (currently non-existent) small set of power users need this feature for their workloads.

There was a lot of discussion in Slack about the best approach to solving this issue, which I am linking to here: https://mitdsg.slack.com/archives/C076WBNJJAH/p1737222894617139