Add GSP shape to dataset

[JackKelly]

When the model is predicting PV yield for an entire Grid Supply Point region, it may be useful for the model to get a binary mask showing the geographical shape of the GSP region.

Here's a handy plot of GSP regions from Dan:

When this mask is given to a fully attentional model, it might make sense to encode the mask as part of the positional encoding for each pixel of satellite imagery. That is, each pixel of satellite imagery would include the usual sin and cos positional encoding for spatial position and temporal position; and perhaps an extra 'position encoding dimension' would be a binary flag to say 'this pixel is (or is not) in the GSP region'.

[JackKelly]

As a first approximation, I wonder if it might make sense to just give our ML models the size of the target GSP region? e.g. give the model the area of the GSP region? (I don't know for sure but I would imagine the geopandas can compute the area in one line of Python)