Add spatial_hash implementation

[nahueespinosa]

For different parts of the algorithm we will need to mark particles that are close to each other with a cluster id. This is used in adaptive sampling to compute the KL-distance and it is also normally used in the estimation process, where they take the most likely cluster (by adding up the weights) and computing the mean and covariance of those samples.

One way of doing that is to discretize particle positions with respect to an user-defined cell size, and then compute a hash that represents that region of space. This is known as voxel hashing.

Here is a super simple implementation that takes a tuple of doubles and has the following properties after rounding the real values:

• In 3-dimensions, it guarantees no collisions in the integer range that goes from -1,048,576 to 1,048,575 per axis (21 bits).
• In 2-dimensions, the usable range goes from -2,147,483,648 to 2,147,483,647 (32 bits).

[hidmic]

@nahueespinosa Awesome work as usual. Left some comments.

[hidmic]

This LGTM if @camistolo is on board.

[nahueespinosa]

Alright, going in!