add warning when trying to create huge VRT

[vincentsarago]

ref #348

Still not sure how to resolve this issue, but for now here I added a warning when the output resolution is bigger than the small overview resolution.

The main problem is that in this case, rio-tiler will create a Huge VRT (matching the input bounds) and use out_shape https://github.com/cogeotiff/rio-tiler/blob/master/rio_tiler/reader.py#L101 to fetch the data for the right overview. But when the output resolution is > to the min resolution for the COG, we will create a bigger VRT and still read part like 256x256 while in reality with might need only 10x10 pixel resampled to 256x256.