Merge branch 'scygan_image_reinterpret_snorm_magic_0x80' into 'master'

Merge branch 'scygan_image_reinterpret_snorm_magic_0x80' into 'master'

dEQP-VK.image.reinterpret: bad handling of -128 SNORM value

Some tests in reintepret group create images of R32I type, than issue an imageLoad()-ed in shader, reinterpreting the image as RGBA8_SNORM. Than the value is imageStore()-ed again as RGBA8_SNORM.

Effectively each channel undergoes SNORM8 -> float -> SNORM8 conversion.

The problem is test requires the stored image to match exactly the loaded one, while the loaded image contains bytes of -128 value.

-128 (same as 127) converts to -1.0 float, however according to following equation in spec, the -1.0 float value can convert only to 0x81, what makes test fail.

```
f′=convertFloatToInt(f*(2^(b−1)−1),b)
```

The proposed fix here is to replace all 0x80 values in loaded image - same it is already done for NaNs and Infs for float format.

See merge request !233