Methods for working with distributions support added for univariate distributions. Please review carefully (repetetive work, easy to make mistakes).
Wasn't sure about EdgeworthAbstract distribution.

`posterior_sample` to update an observation model from a posterior sample.

This is useful for generic algorithms that work on arbitrary types of
distributions (e.g. fitting a finite mixture model).

fallback rand was catching distributions with rand undefined (e.g. Kolmogorov), returning method_exists as true, and triggering a stack overflow when called.

For many distributions the support does not depend on the parameters of the distribution. This adds methods for
```julia
insupport(Normal, x)
```
in addition to
```julia
insupport(Normal(), x)
```

Also many of the checks in constructors were condensed to the
```julia
check || error("check failed")
```
or
```julia
check ? new(args...) : error("check failed")
```
form.

Some external constructors were moved to internal constructors when possible.

Checks of the form `x > 0.0` were rewritten as `x > zero(x)` so that `x` is not converted before the check.

Basically, I moved things around to make them more organized -- related
functions are put together.

In addition, all vectorized functions for univariate distributions are
generated using macros in a consistent away.

In addition, there are minor tweaking of other functions.

The refactor has passed tests.

Add EmpiricalUnivariateDistribution

* Added more entropy() methods
* Added tests for entropy based on the AEP
* Fixed a bug in the fallback method for pdf!()