Fix get_terrain_attribute bug with edge_method='none'

[rhugonnet]

Resolves #244

• Fixes an indexation issueof get_terrain_attribute with edge_method='none',
• Removes wrong referencing of slope equation in comments,
• Changes the default edge behaviour to match that of GDAL (i.e. edge_method='none' and fill_method=none`),
• Adds doctests for slope convenience function, and adapt doctests of other attributes with new default methods.

We should have more specific, in-depth tests for the edge and filling methods. Not the aim of this PR, so opening an issue for it.

[erikmannerfelt]

Thanks for improving on these methods!

It's uncontroversial to default to the GDAL functionality, but I think GDAL is wrong in propagating NaNs like that. If you think we should default to GDAL-behavior, then sure.

[rhugonnet]

Thanks for improving on these methods!

It's uncontroversial to default to the GDAL functionality, but I think GDAL is wrong in propagating NaNs like that. If you think we should default to GDAL-behavior, then sure.

When you are missing only one point in the corner, I agree the results is still OK! But when you are missing half of the cell, the terrain attribute value will be aberrant.
Maybe an argument could control when the filling method is applied, depending on the amount of gaps to fill in the 3x3 window? Then I would certainly be fine having a default method that fills when there's only one missing pixel for example 😉
With the current approach, we should definitely default to GDAL to avoid all those pixels that end up with erroneous estimates (usually we have enough pixels in DEM data anyway!)

[erikmannerfelt]

Great!