Regarding the first question, my understanding is that when you run 3dDeconvolve you don't need to explicitly specify anything about the trend in your model no matter it is for regression or deconvolution, for event-related experiment or block design type. The program uses those timepoints that are not explicitly expressed in the stimulus files for trend parameter estimation with the type of baseline specified by the user (the default is straight line). And the baseline is modeled separately regardless of the lags in the deconvolution.
Gang