bugfix: handle problems acos arg > 1

[tjansson60]

While processing the shoreline set GSHHS_c_L1 with 742 shapes a
problem was encountered where cos_value = 1.00000000000048 which meant
that the acos raised an ValueError: math domain error and the graph
could not be built.

This patch checks if the arguments are very close to -1 or 1 and returns
the expected values. If the values are not close to either of these
values the ValueError is still raised.

Examples of values in angle2 when problem was first encountered:
point_a = (112.96, -25.49)
point_b = (45.00, -25.49)
point_c = (45.00, -25.49)
a = 6.938890000001394e-07
b = 4618.002849783456
c = 4618.11606498842
cos_value = 1.00000000000048
T = 10000000000000
T2 = 10000000000000.0

[TaipanRex]

I'll need to performance test this.

Check out the discussion on angle2() in #28, by setting COLIN_TOLERANCE to a lower value, like 8, it should solve this problem without any performance hit. Its not explained well in the code and I'm working on better naming the COLIN_TOLERANCE, T1, T2 variables and also allowing the user to set COLIN_TOLERANCE through the API (see #29).

[tjansson60]

Ahh, had not seen #28, but yes it the same. I think it should not hit performance with a try except loop as it only executes the except if the try fails, so I am assuming it should be be somewhat the same. I guess this ties together with your work on a performance testing framework in #30.