PF in-the-loop and convergence failure

With multi-period power flow in the loop, how should we handle non-convergence?

Right now, solve_power_flow! in power_flow_evaluation.jl sets pf_e_data.is_solved to true, even if some (or all) time steps failed to converge. PowerFlows.jl writes NaNs to certain fields when the power flow fails to converge. Side effects:

• The NaNs in the bus_angle and bus_magnitude fields infect power flows in subsequent simulation steps: PSI re-uses the same PowerFlowData object and clear_injection_data! only zeroes out the power-related fields.
• calculate_aux_variable_value! on the power flow related variables produces NaNs. The code fetches the values from the most recent simulation step for which is_solved == true, but that's all of them.

PSI has a single variable: pf_e_data.is_solved. PF has pf_data.converged::Vector{Bool}, with one true/false per time step. What should be the relationship between the two? is_solved = all(converged)? is_solved = any(converged)?