Fix nlmeans indexing

[samuelstjean]

I though I had made an issue about it, must have been a mental note.

Anyway, the cython part of nlmeans had a few indexing error which inverted along the way the first and second dimension, so hopefully I fixed them here. That's why not all the code is changed, the main loop was using j,i,k, but passing i,j,k down, so now they should match.

I also revamped the calling file to remove parts I personally deemed unnecessary at the same time. From memory, the example also had some statements which were not truc, so I'll have a look afterwards if this here is fine.

[samuelstjean]

There it is after all #1178. It fixes half the stuff from the issue (as per title), the other half seems to be fixed around in #1245

[samuelstjean]

erm are they fixing it in the other pr? it seemed like nobody understand what I meant, half the indexing in the other loop works, and the one in the inner loop is different. It only has a small effect since it swaps local coordinate, but it is still wrong, that's why you only need to change half of them.