(dbscan) Use valid index for neighbors

[ezheidtmann]

The DBSCAN implementation was producing garbage results for me because the _get_neighbors function was returning the wrong indices for valid neighbors.

If comparing a sample with higher index than the argument, the returned indices were off by 1.

This change fixes it.

PS thanks for this implementation! It's a better fit for my use-case, where speed is not important but avoiding scipy is very convenient.