Make the decoder more robust against extra trailing bits in the NALUs

[DCNick3]

I initially implemented this as a solution for #106. While my issue turned to actually be not-a-bug, it seems that it still exists in the wild: https://github.com/scottlamb/retina/issues/102.

It mimics the logic of ffmpeg for reading rbsp_trailing_bits, while issuing a warning in case of extra trailing bytes.