It seems to me that your understanding is correct. Here are just a few thoughts from a different perspective:

(1) The impulse response function in a time-invariant linear system shows how the output signal intensity is related to the input as demonstrated in the corresponding convolution integral or its discrete counterpart. The name 'impulse response' reflects the fact that this function describes the system's response to an impulse input of unit size. Thus in the response domain the IRF also plays a role of unit size as the impulse does in the input domain.

In fMRI world with output signal and stimulus available, we estimate the IRF through back-engineering, solving the convolution integral or realistically its discrete form.

(2) Partial F corresponding to a stimulus or regressor would tell how statistically significant this stimulus or regressor is accounting for the whole variation in the output signal.

(3) Full F tests whether all regressors (or stimuli) are statistically significant, that is, is the signal at a voxel just some noise or activation and noise?

(4) minlag and maxlag specifiy both delay and duration of the IRF.

Gang