Fix preferred encoding computation

[nyurik]

• Decides which encoding to use depending on both Accept-Encode and the preferred encoding (chooses the preferred one in case brotli and gzip have the same q value)
• Uses gzip by default

This will fix #1315

[danielrh]

I've measured a lot of brotli vs gzip comparisons, and have always found that with the appropriate settings brotli is faster to compress, decompress and results in smaller filesize at every comparable setting.

It sounds like something is amiss with some setting that would cause gzip to ever win in a fair comparison--it's just an older and less capable algorithm.
What was being measured here?