Implement likelihood estimation for particles

[nahueespinosa]

Following up on #24, this patch implements the likelihood estimation step and a stationary motion model.

Related to #3, as usual.

Peek.2022-12-06.23-03.mp4

Peek.2022-12-06.23-04.mp4

[nahueespinosa]

Pending items (probably not for this PR):

• Update policy for the particle filter: the update happens on the laser scan callback and we publish particle clouds every 500 ms. Not sure if that will be the final solution.
• Generic state representation.
• The sensor model is still part of beluga_amcl but should probably be moved to the core package.
• Real motion model.
• The model used in the flatland simulation doesn't respond well to teleop_twist_keyboard controls (might be my latin american keyboard configuration though).
• nav2_amcl details and hacks that were not re-implemented here:
  • Ad-hoc weighting scheme.
  • Combining weights from previous sensor readings between filter updates?
• Extensions to the likelihood computation mentioned in Probabilistic Robotics 6.4.2:
  • Sort map occupancy values into free, occupied and unknown instead of just the first two. Useful if the map contains unexplored areas.
  • Match a scan and a map symmetrically: In addition to calculating the likelihood of a scan field inside the map, a symmetric extension also calculates the likelihood of each (nearby) object in the map object relative to the likelihood field of a scan.
• Estimated pose computation and publishing.
• Model tests.

[nahueespinosa]

[hidmic]

First pass