Your model is wrong. Using model_diagnostics returrns
MODEL_DIAGNOSTICS: The following endogenous variables aren’t present at the current period in the model:
k
On top of that, your model seems to have an improperly handled constant term in equation 19. Due to linearization, your model variables should all have a steady state of 0. But your Taylor rule contains a constant term pi_target that should not be there in this form. This causes the weird steady state.