[Feature Request]: A safe single threaded pipeline engine

### What would you like to happen?

The pipeline engine would execute every transform one by one, never in multiple copies.
The rows are serialized to disk after every transform execution in a safe, configurable, location.
When the pipeline fails it should be possible to simply execute the pipeline with the same again and it will continue with the transform that failed the last time around.

Most of the code already exists. We essentially only need a spooling rowset implementation of `IRowSet`.

### Issue Priority

Priority: 3

### Issue Component

Component: Pipelines

Start by locating the IRowSet interface and the existing pipeline engine implementation, then review how transforms and rowsets are currently executed. The work is done when a configurable spooling rowset serializes rows after each transform, runs transforms single-threaded, and resumes from the failed transform after rerunning the pipeline.

Written by the indexing model from the issue text.