[Bug Report] Phash validation

[kermieisinthehouse]

Describe the bug
We recently started using phashes for matching against StashDB. This has exposed bugs in the phash generation process. A broken or unparsable file will result in common phashes, which will then be matched to a random scene in StashDB that has had that same broken phash uploaded. We may not be able to fix the underlying ffmpeg issues, but we can work around them. During phash generation / use, we should validate to make sure the phash doesn't match a known bad phash ( results of solid color, color bars, etc.), and other phash validation rules.

Further investigation is still needed, but I've identified some facts already:
Real phashes should have a (roughly?) even amount of 1s and 0s
Many bad phashes 'look' strange. They have slow entropy.
If ffmpeg determines that the video duration is zero, the phash is almost always junk.

Known bad phashes so far:
(note, they may vary in the wild by 1-3 bits, so checks should check for a hamming-distance match)

a000000000800080
8080808080808080
870707030787fefc
82070707078ffff8
8055557555575575
805555555d755d55
87070707037ef8fc
8707070303fefcdc
cdcdcdc9c1233332

[kermieisinthehouse]

Also,

f8fcfcfcf8e00100
f8fcfcfcfc700000