Add explicit guide for truncated distributions

[eirikbrandsaas]

It wasn't obvious that I could truncate the distributions so I added that to the documentation, to the ArgumentError, and a test.

Somewhat related to #40

[eirikbrandsaas]

While I did add tests, I didn't see any change in the number of tests when I ran ] test locally. Don't know what's up with that.

[eirikbrandsaas]

Or maybe this wasn't a great idea:

E = expectation(truncated(Pareto(),0.0,1000.0))
sum(E.weights) # ≈ 1.0

E = expectation(truncated(Pareto(),0.0,1000.0),n=50)
sum(E.weights) # ≈ 0.97

Or maybe it isn't a problem that the weights don't sum to one for the algorithms in expectations?

[arnavs]

@eirikbrandsaas Thanks for this!

Not sure what's up with tests. I'm guessing that the package manager wasn't actually testing the changed package for some reason.

I think the weights should sum to 1, but I'm more flexible when the user is specifying the number of nodes instead of using the defaults.

[arnavs]

(i.e., drawing 500 nodes once might still be an improvement over repeated Monte Carlo sampling, and drawing 50 nodes once might be acceptable with "3% error")