Before we'd round up all preview request to their nearest power of two.
This resulted still in a lot of possible images. Generating a lot of
server load and taking up a lot of space.
This moves it to previews to be powers of 4: 64, 256, 1024 and 4096
Also the first two powers are always skipped (4, 16) as it doesn't make
sense to generate previews for that.
We cache preview pretty agressively and I feel this is a better
tradeoff.