Propogation of lead and lag to downstream library

[JimCircadian]

The propogation of lead and lag to the end of the processing chain is still giving me awkwardness. This is primarily dealt with by preprocess_dataset and the commands herein. We should be in a position where:

1. Datasets have the full date range assigned
2. When splits are provided, the split coverage is adjusted based on available data
3. When loaders are produced, they only contain the dates that can be covered by the data taking into account lag and lead

Capturing the train of thought through the configuration assets to give myself some confidence this is working properly, before changing the implementation at the end of the line (where icenet is now having this behaviour documented properly.)