Add rate matrix estimation

[franknoe]

I am just copying this issue from PyEMMA. Probably the text below is outdated though.

@fabian-paul , whenever you feel confident about the robustness of your rate matrix estimation, please commit a separate feature branch. We will include it in the 1.3 release (I will do the merge).

I suggest to do this as follows:

1. Add a new method to pyemma.msm.estimation. I'm not sure exactly what the input is in your case, but I imagine something like

   def rate_matrix(dtrajs, tau, use_connectivity=True, reversible=True)
      """Estimates a rate matrix from discrete trajectories by likelihood maximization at lag time tau.
   
      Parameters
      -------------
      dtrajs : int-array or list of int-arrays
          discrete trajectories
      tau : int
          lag time at which the likelihood is evaluated
      use_connectivity : bool, optional, default=True
          If true, only the rates with c_ij(1)+c_ji(1) > 0 will be estimated. 
          If false, all rates with c_ij(tau)+c_ji(tau) can be nonzero
      reversible : bool, optional, default=True
          If true, the estimated rate matrix is the maximum likelihood reversible rate matrix.
      """

   This is a little different from the other estimation method in msm.estimation, which do an elementary step (trajectories -> count matrix; count matrix -> transition matrix). Perhaps the same can be done here, I am not sure.

2. Add a new property rate_matrix() to the EstimatedMSM object which calls

   rate_matrix(self.active_dtrajs, self.lag, use_connectivity=True, reversible=self.is_reversible)

[marscher]

fixed by #57