> 1. I am assuming it is appropriate to use preprocessed data (with some
> detrending/filtering/normalization already applied) - is this correct?

Yes, that is appropriate in creating the interaction regressors.

> 2. Is the 3dSynthesize/3dCalc method now recommended in place of 3dDetrend?

Yes I believe the 3dSynthesize route is more preferable.

> 3. To confirm, If 3dSynthesize is preferred, should one still include the
> motion and nuisance regressors in the final 3dDeconvolve step even
> though they are included in 3dSynthesize/3dCalc?

That's correct because the previous step with 3dSynthesize/3dCalc is only for creating interaction regressors. Those confounding effects would still be present in the original data (input for 3dDeconvolve).

> 4. A bit off topic - when conducting timecourse analyses (or simply
> plotting averaged timecourses such as % signal change) is it
> advisable to first use 3dSynthesize/3dCalc? Or is the original
> preprocessed data better?

Yes it is a good idea to remove those confounding effects as much as possible when showing the task-related effects.

Gang