Use k-NN to compute ego-centric road observations

[SamanKazemkhani]

This introduces selectKNearestRoadEntities(), which computes a given agent's ego-centric observations of road entities. It performs the following sequence of steps to fill in AgentMapObservations:

1. Find the K nearest road entities to the agent, where K is defined by consts::kMaxAgentMapObservationsCount
2. Filter out road entities further than observationRadius computed in (1)
3. Zero out values if there are fewer than K road-entity observations left

(1) is accomplished by using a binary heap (implementation in binary_heap.hpp) with time-complexity O(N * log K) and space-complexity O(K).