Add a built-in neighbor list "filter" for tuning functions

[GardevoirX]

Suggested by @ceriottm, in cases where users want to optimize the cutoff, it would be nice to have a built-in "filter", so the user can do the NL calculation with a large cutoff for only once, and then reuse the result during the optimization. Along with this update, new tests shall be designed and added.

For reference:

torch-pme/examples/10-tuning.py

Lines 205 to 210 in 6daacf3

	def filter_neighbors(cutoff, neighbor_indices, neighbor_distances):
	assert cutoff <= max_cutoff
	filter_idx = torch.where(neighbor_distances <= cutoff)
	return neighbor_indices[filter_idx], neighbor_distances[filter_idx]