padding

[keewis]

As part of #18, we need to pad the input array. #23 added the basic infrastructure and the constant mode.

Planned modes:

• [x] constant: fill padded cells with a constant value
• [ ] edge: fill padded cells with the closest edge cell ("nearest neighbour interpolation")
• [ ] linear_ramp: linear interpolation from the closest edge cell to a constant value
• [ ] symmetric: reflect the values along the edge of the domain
• [x] local aggregation of the neighbourhood: mean, mode, max, min
• [ ] scale aggregation: mean, mode, max, min (this refers to the second mode from https://github.com/IAOCEA/healpix-convolution/issues/25#issuecomment-2214375365, I just can't find a better name for this)

The remaining cells all need a way to find the nearest neighbour from the list of computed neighbours. In addition to that, linear_ramp needs the distance in rings, and symmetric needs to work with vector geometry to find the right cells.

[keewis]

as pointed out by @jmdelouis, it looks like for some cells there is multiple nearest neighbours, so we'd have to compute the average of those neighbours.

Instead, we have a couple of other modes: mean, mode, max, min (simple aggregation of the neighbourhood – group neighbours by the padded cell to implement this), and a different mode (mostly mean, but could also be median / max, min) that includes more data the further the padded cell is away from any data cells to capture larger-scale features.