Resume failed process

**Is your feature request related to a problem? Please describe.**
When a process (`map`, `filter`, etc.) crashes part-way through, you lose all progress.

**Describe the solution you'd like**
It would be good if the cache reflected the partial progress, so that after we restart the script, the process can restart where it left off.

**Describe alternatives you've considered**
Doing processing outside of `datasets`, by writing the dataset to json files and building a restart mechanism myself.

**Additional context**
N/A