Follow-up from "h264decoder: Don't handle gap frame num for the first picture"

The following discussion from !1204 (merged) should be addressed:

• @seungha.yang started a discussion: (+3 comments)

  @ndufresne This MR doesn't fix the issue for nvidia stateless codec implementation so I guess v4l2 stateless decoder is most likely the same case. but d3d11h264dec (both intel and nvidia) looks fine with this MR. So weird...