Cleanup Degrain AVX, similar to SSE2 code.

There is no major change in performance. The new code is *slightly*
slower, but we're talking very small differences here, like maybe ~1%.

Tests run on blocksize 8 and 16.

Here are some numbers for 540p and 1080p. Note the `mvversion` arg for
oldavx vs newavx:

```
Summary
  'vspipe -p -e 3000 -o 2 --arg mvversion=oldavx --arg src=test-540p.dgi tester.vpy /dev/null' ran
    1.00 ± 0.00 times faster than 'vspipe -p -e 3000 -o 2 --arg mvversion=newavx --arg src=test-540p.dgi tester.vpy /dev/null'
    2.69 ± 0.01 times faster than 'vspipe -p -e 3000 -o 1 --arg mvversion=oldavx --arg src=test-540p.dgi tester.vpy /dev/null'
    2.70 ± 0.00 times faster than 'vspipe -p -e 3000 -o 1 --arg mvversion=newavx --arg src=test-540p.dgi tester.vpy /dev/null'
    3.14 ± 0.01 times faster than 'vspipe -p -e 3000 -o 2 --arg mvversion=oldavx --arg src=test-1080p.dgi tester.vpy /dev/null'
    3.15 ± 0.01 times faster than 'vspipe -p -e 3000 -o 2 --arg mvversion=newavx --arg src=test-1080p.dgi tester.vpy /dev/null'
    8.88 ± 0.04 times faster than 'vspipe -p -e 3000 -o 1 --arg mvversion=oldavx --arg src=test-1080p.dgi tester.vpy /dev/null'
    8.89 ± 0.02 times faster than 'vspipe -p -e 3000 -o 1 --arg mvversion=newavx --arg src=test-1080p.dgi tester.vpy /dev/null'
```

Same thing for 4k:
```
Summary
  'vspipe -p -e 500 -o 2 --arg mvversion=oldavx --arg src=test-4k.dgi tester.vpy /dev/null' ran
    1.01 ± 0.00 times faster than 'vspipe -p -e 500 -o 2 --arg mvversion=newavx --arg src=test-4k.dgi tester.vpy /dev/null'
    2.69 ± 0.00 times faster than 'vspipe -p -e 500 -o 1 --arg mvversion=oldavx --arg src=test-4k.dgi tester.vpy /dev/null'
    2.69 ± 0.00 times faster than 'vspipe -p -e 500 -o 1 --arg mvversion=newavx --arg src=test-4k.dgi tester.vpy /dev/null'

```

Version 23

Version 22

Version 21

Version 20

Version 19

Version 18

Version 17

Version 16

Version 15