Improves benthamHalfLookaheadValueOfCell() efficiency by ~40%

[colinhanrahan]

In my test using benthamHalfLookaheadTop, 75% of findBenthamHalfLookaheadValueOfCell() runtime is taken up by one call to agent.findNeighborhood(cell) which is then passed to len(agent.findNeighborhood(cell)). 80% of the call to agent.findNeighborhood(cell) is self.findCellsInRange(newCell), and 60% of that is shuffling. For this specific use case, there is no need to shuffle — you just want the size of the neighborhood at the next cell.

findNeighborhood() is called 4 times in the repository:

• at the top of findBestCell(), with no parameter, so newCell == None. This changes self.neighborhood so it should shuffle before returning.
• once in each *HalfLookaheadValueOfCell() for a total of 3 calls, with a cell parameter, so newCell != None. The return values from these 3 are only used to find a length, so the lists do not need to be shuffled.

findCellsInRange() is only used for findNeighborhood(). I've moved the shuffle() from findCellsInRange() to the "if newCell == None" block in findNeighborhood(). This will also improve all other *HalfLookaheadValueOfCell() functions by a similar percentage, but I didn't test them all individually. In my testing, the change seems to speed up benthamHalfLookaheadTop total simulation runtimes by 30%.

[colinhanrahan]

Removed shuffle and comment in new commit.

[nkremerh]

Such a small change netted ~5% to ~45% improvements in simulation runtime using the Bentham decision model and did not cause any catastrophic changes in overall agent population (numbers are relatively stable on average across a handful of seeds). Fantastic catch!