Consider measurement error in the forward simulations

[cboettig]

The GP assumes measurement error, should probably be comparing to a forward simulation with measurement error. Likewise should have the optimal parametric solution account for the measurement error, if not the alternate one.

[cboettig]

A related issue: The parametric estimation should use the estimated sigma_g from the MLE in place of the real one.

[cboettig]

Including measurement error should modify not only the simulation but the policy calculation though, see multiple_uncertainty tag in pdg_control.