Memory efficiency when treating lines as points

[eoydvin]

Currently it is possible to treat CMLs as points by passing full_line=False. This simplifies the distance matrix in bk_functions before performing block kriging by doing

         # Force interpolator to use only midpoint
         if full_line is False:
             x0 = x0[:, :, [int(x0.shape[2] / 2)]]

However, for large CML datasets x0 becomes very large, requiring more memory than what can be handled on a personal computer. In the future, setting full_line = False, should treat CMLs as points already when constructing x0.