EHN: Add qhull_options to game_theory.vertex_enumeration

[oyamad]

For some degenerate games, qhull_options='QJ' helps to identify all the Nash equilibria:

bimatrix = [[(0, 3), (6, 1)],
            [(2, 2), (5, 6)],
            [(3, 3), (3, 3)]]
g = gt.NormalFormGame(bimatrix)

gt.vertex_enumeration(g)

[(array([0.66666667, 0.33333333, 0.        ]),
  array([0.33333333, 0.66666667]))]

gt.vertex_enumeration(g, qhull_options='QJ')

[(array([0., 0., 1.]), array([1., 0.])),
 (array([0.0000000e+00, 3.9409863e-11, 1.0000000e+00]),
  array([0.66666667, 0.33333333])),
 (array([0.66666667, 0.33333333, 0.        ]),
  array([0.33333333, 0.66666667]))]

but not always:

g = gt.NormalFormGame((2, 2))  # payoffs all zero

gt.vertex_enumeration(g)

[(array([1., 0.]), array([1., 0.]))]

gt.vertex_enumeration(g, qhull_options='QJ')

[(array([1., 0.]), array([0., 1.]))]

[coveralls]

Coverage increased (+0.01%) to 95.117% when pulling 463d05e on qhull_options into 4b94234 on master.