committed vectorized normal_log() and special function square()

[Bob Carpenter]

I cleaned up some of the implementation of the
univariate log_norm by adding a special function
square() and code to do its auto dif.

I also added a vectorized log_norm(). I overloaded
log_norm, so that it'd be OK with:

normal_log(vector<T_y> y, T_loc mu, T_scale sigma);

as well as the original

normal_log(T_y y, T_loc mu, T_scale sigma);

We already have a separate multivariate_normal, so let
me know if you object to this overloading.

- Bob

[Daniel Lee]

That looks good to me.