CDO

Hi all,

I would like to remap (128 x 128) regular grid in Lambert projection to a regular grid in latlon projection. I am perfectly able to do it using `cdo remapbil`.

However, I would like to incorporate it into my Python pipeline, so I would like to use `cdo genbil` and then work with the weights in Python like:
```
dst_add = weights["dst_address"].to_numpy()
src_add = weights["src_address"].to_numpy()
map_wts = weights["remap_matrix"].to_numpy().squeeze()

for n in range(int(num_links / 4)):
noff = n * 4

dst_array[dst_add[noff]] = (
        src_array[src_add[noff]] * map_wts[1 * noff] +
        src_array[src_add[noff + 1]] * map_wts[1 * (noff + 1)] +
        src_array[src_add[noff + 2]] * map_wts[1 * (noff + 2)] +
        src_array[src_add[noff + 3]] * map_wts[1 * (noff + 3)]
    )
```

This seems quite straightforward, but for some reason I am getting a substantially different result (see attached file).

I suspect that I am missing some operation performed on the input grid, but I could not find it in the CDO code myself.

Please advise.