A star: functional programming with numba

[thuydotm]

• separate snap option as snap_start and snap_goal
• add connectivity option, can be either 4 or 8, defaulted as 8
• warning if start or end at a non-crossable pixel (nan or a barrier)
• raise error if start or goal location outside the surface graph
• benchmark and numba-ize
• add Pathfinding Supported Spatial Functions in README.md
• add example to user_guide.ipynb

[thuydotm]

@brendancol Here the benchmark of A* pathfinding with the example of the city of Austin road network

# street_agg.shape = (600, 800)
%timeit path_agg = a_star_search(street_agg, start, goal, barriers=[0])

# 1. Old version using PriorityQueue:
11min 44s ± 21.5 s per loop (mean ± std. dev. of 7 runs, 1 loop each)

# 2. New version of functional programming with Numba
160 ms ± 32.1 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

The new version is ~3500 to 5700 times faster.

[brendancol]

@thuydotm Wow!! So great!