Revisit interpolation classes and Ocean class

Extrapolation

• Warning should be raised when user tries to get data outside area/volume with data support.

• Revisit how Kadlue handles extrapolation outside area/volume covered by data points. Nearest neighbour would appear to provide the safest (default) choice. The only drawback would be potential issues with discontinuities when modelling sound propagation.

Interpolation

• Choice of interpolation method should be exposed via Ocean class constructor, separetely for each data type
• Time averaging should not be applied as default to data with temporal dependency (e.g. temp, salinity, etc.), but could still be an option

Loading of data sampled on irregular grids:

There are several issues with how Kadlu currently handles loading of data sampled on irregular grids:

• if passed as numpy arrays, the Ocean class attempts to create a regular grid (using some implicit and not very well documented rule for replacing missing values) which leads to memory problems for large input arrays; instead the Ocean class should pass the arrays as are to the Interpolator which already has a solution implemented for handling large irregular grids

• the rules for replacing missing values are not well documented, and should in any case be made more configurable