improve performance of AndNot between array and run containers

[neena]

Improve the performance of in-place AndNot between array and run containers. The current approach converts both to bitmaps, performs the AndNot, and then returns the result.

Instead, this approach scans the list of runs in the run container, and calls iremoveRange on each one in the array container.

Before:

BenchmarkAndNot/inPlace=true/left=array/right=run-10        	  121395	     11533 ns/op

After:

BenchmarkAndNot/inPlace=true/left=array/right=run-10       	 1308555	      2369 ns/op

[lemire]

Tell me what you think of #417

I think that 417 might be faster.