Fully order the pvq search candidates for portable, stable sorting

[barrbrain]

I observed different results from AWCY when running the same commit, test clip and configuration on my MacBook. It occurred to me that it could be a result of differences in qsort implementation. Adjusting the sort key to produce a stable sort indeed fixed this for me.

https://arewecompressedyet.com/?r%5B%5D=pvq-sort-good-2016-09-06T02-43-53.523Z&r%5B%5D=master_2016-09-03d&s=objective-1-fast

[barrbrain]

Added a comment and a better commit message.

Strictly speaking, the natural stable order is fully specified by breaking ties with (gain, theta). Gain was sufficient for my test but whether that is absolutely true depends on how k is computed.

[jmvalin]

r+

[barrbrain]

Thanks for the review. Any plans to merge this?