> 1. Is it necessary to model all the stimuli (i.e., have 4 regressors in one 3dDeconvolve command)?

Yes, whatever you think have some signature in the BOLD signal should be accounted for in the model.

> 2. What is 3dDeconvolve using as the “baseline”?

The baseline condition is considered at those time points when fixation or none of the condition are present, and is modeled by default with a horizontal line (plus slow drift through polynomials).

> a. For example, if I do put all 4 regressors in – and I look at the activation associated with condition A,
> is this activation during condition A vs. the activation during any timepoints that are not modeled by any
> of the 4 regressors?

Yes, that's right.

> b. If so, this would make me think that you need to model all 4 stimuli because if you only modeled 2,
> then the 2 conditions not modeled would be considered part of the baseline. Is this correct?

Yes, correct.

Gang