Added baseline statistics for Cox

[piever]

This is part 4 of #5 . I've added three functions to get baseline hazard, cumulative hazard and survival functions of a fitted CoxModel and added some docs.

[piever]

I'm not sure what should baseline_survival (and similar) return: now it gives a pair of vectors (times and values), but maybe it should give a function that can be tested on any new value, like ecdf? Or maybe even both and we could create new type that has both x and y fields (with these vectors) but is also callable. What do you think?

[nalimilan]

I agree it makes sense to return a special distribution object. I think there have been discussions about how to represent empirical distributions in the same framework as the Distributions packages. @andreasnoack knows better.

[piever]

That'd be very helpful. I should probably leave as is here for now and maybe we can open an issue about this at StatsBase (where I think this kind of object should be defined)? I think the new type should have x and y field for the specific points, a f field with the function and potentially a few constructors to simplify building f in the common cases.

[nalimilan]

I guess the type would be defined in this package, but it could inherit from and implement an abstract type defined e.g. in Distributions.