The DOS computed for a device configuration is indeed the device DOS, but it's computed for the atoms in the central region (with device boundary conditions, not bulk), i.e. the electrodes (which are bulk) are not included.
The magnitude of the DOS depends quite strongly on the k-point and energy sampling. The peaks are often quite sharp, so when they move a little bit due to the gate, your sample point (in energy) may end up on different points of the peak. The effect should go away to some extent with a higher k-point sampling and denser energy points, however not completely, it is also expected that there is some changes in the electronic states due to the gate, so part of the amplitude can be due to that.