For a very long time, OPS has had the ability to use a Gaussian bias in shooting point selection. That is, given a CV f, the probability of selecting a shooting point x within a given trajectory is proportional to np.exp(-a*(f(x)-x0)**2), where a and x0 are parameters set by the user.

There is no reason for this to be restricted to Gaussians. This PR introduces a BiasedSelector, which allows one to use any function (wrapped in an OPS CV) as the bias function. The only reason this didn't exist before is because when we were writing shooting point selectors, I don't think we understood the flexibility that OPS CVs would have.

Now, for example, the GaussianBiasSelector could be implemented as:

def gaussian(snapshot, cv, a, x0):
    return np.exp(-a * (cv(snapshot)-x0)**2)

cv_gaussian = paths.FunctionCV('shooting_bias', gaussian, cv=my_cv, a=2.0, x0=0.25)
selector = paths.BiasedSelector(cv_gaussian)

But similarly, one could use any other function.