You do not need the model(linear), but if your model is linear, it speeds up computations a bit.
You should not use the prefilter option, because both your data and and your model variables are mean 0 (see also “A Guide to Specifying Observation Equations for the Estimation of DSGE Models”)
Did you check what happens if you take Eviews’s final values as starting values in Dynare? Is the likelihood higher or lower than with the values found with Dynare?