#7893 fixed a case where copies into the staging buffer could read from uninitialized memory. In discussion on that PR, it came up that we probably should clear any padding area in the staging buffer that will not be populated from the source of the transfer.