Starting with with a model where no correction is needed, the coefficient on the one-period lag gives you the speed of adjustment.Suppose you obtained -.30 and your data are annual and natural log was used. Then, it implies that 30 percent of the movement of the movements in your dependent  variable is attributed to the difference between actual and the equilibrium levels in a given year.Put differently, 30% of. The adjustment occurs in a year.

Sorry, I forgot your second question. I will try to add more lagged dependent variables to remove the serial correlation. I would attempt to use AR(1). It is never a good research procedure when you use AR(1) or more corrector.

First of all, ARDL model is used for removing auto-correlation from a series either I(1) or I(d) by including lags of dependent and independent variables to whiten the error term of the model. It is one of the dynamic model used for observing short run relationship between variables. As far as your question is concerned, AR(1) means dependent variable depending on its first lag can be used to remove auto-correlation if there exist first order auto-correlation in the model an it is AR(1) but not the ARDL model because it contain only the lag of  dependent variable not the independent variable.

Absolutley, there may be complex long term relationships that do not follow a simple AR(1) process