Implement line sampling for failure probability

[FriesischScott]

This adds LineSampling and the required methods such as computation of the gradient and mapping to standard normal/physical space. New dependencies are:

• FiniteDifferences.jl to calculate gradients
• Dierckx.jl to interpolate and find the root on the line when calculating the pf with LineSampling

I also introduced two new types for deterministic and random inputs. This allows us to better filter for random/ non-random inputs.

In addition, I switched all names from ::String to ::Symbol to make working with the DataFrames easier.

I'm not 100% sure about some of the method names. Especially for the mappings between the spaces.

[coveralls]

Coverage decreased (-5.3%) to 94.737% when pulling cc7985c0998782e2071583b874bbdaa48768d9b8 on line-sampling into 12ae98f8ee339b3c6084b511a1aeb56ba65c1f22 on master.

[FriesischScott]

I am going to merge this now. @AnderGray if you have any remarks/changes you'd like to see, let me know.