Switch line-clipping algorithm

[jbcrail]

I switched from Cohen-Sutherland to Liang-Barsky. The performance gains for random lines range from 50-75% improvement for a million lines.

Related to #456

[jbcrail]

To compare the original (Cohen-Sutherland) and new (Liang-Barsky) algorithms, I generated a random set of lines with variations in how many lines are outside the clipping window:

sample lower-left : (0, 0)
sample upper-right: (10000, 10000)
window lower-left : (0, 0)
window upper-right: (10000, 10000)
iterations : 1000
# points   : 1000000
  cohen-sutherland:  9.2482s
  liang-barsky    :  2.2957s (75.18% faster)

sample lower-left : (-100000, -100000)
sample upper-right: (110000, 110000)
window lower-left : (0, 0)
window upper-right: (10000, 10000)
iterations : 1000
# points   : 1000000
  cohen-sutherland: 29.6957s
  liang-barsky    : 13.4035s (54.86% faster)

sample lower-left : (0, 0)
sample upper-right: (110000, 110000)
window lower-left : (0, 0)
window upper-right: (10000, 10000)
iterations : 1000
# points   : 1000000
  cohen-sutherland: 39.7649s
  liang-barsky    : 15.4240s (61.21% faster)

sample lower-left : (-100000, -100000)
sample upper-right: (10000, 10000)
window lower-left : (0, 0)
window upper-right: (10000, 10000)
iterations : 1000
# points   : 1000000
  cohen-sutherland: 48.8722s
  liang-barsky    : 18.0797s (63.01% faster)

Since the clipping algorithm is always run prior to drawing a line, the performance gains should dominate the overall performance of line().

[jbednar]

Thanks!