Tutorial on burn scar analysis using embeddings from partial inputs

[yellowcap]

This PR adds a tutorial on Burn scar analysis using embeddings from partial inputs

This notebook contains a complete example for how to run Clay. It combines the following three different aspects

1. Create single-chip datacubes with time series data for a location and a date range
2. Run the model with partial inputs, in this case RGB + NIR
3. Study burn scares through the embeddings generated for that datacube

Closes #126
Closes #136

[weiji14]

Thanks @yellowcap, mostly just minor suggestions. I'd re-run the notebooks again with the nearest-neighbour resampling to make sure the results don't change much. Also would be good if @lillythomas can check the PCA part since she's been working on that more recently.

[weiji14]

Should we document the mean/std values for all the bands (not just RGB+NIR) somewhere for v0? At least until we resolve #94?

[yellowcap]

Added docs for this in #152

[lillythomas]

This looks great @yellowcap good to see how you plotted the dates directly from the stack. That works well for single component scenarios, which should be largely sufficient for many stories. More complex ones might require somehting like what I did where I added the non-normalized dates to the batches such that embeddings can be labeled with dates in the plots.

Signal of the burn scars is clear. This looks good to merge. I'll incorporate some takeaways from this into my PR.