[Vulkan][KosmicKrisp] frac.32.test is failing on some of the NaN operations.

[farzonl]

This migt just be us testing undefined behavior. maybe we shouldn't be doing NaN tests for this? test is failing on both clang and dxc.

Bug

- [ nan, nan, 0, 0, 0, 0.5, nan, 0, 0.71828, 0.599976, 0.611, nan ]
+ [ nan,   0, 0, 0, 0, 0.5,   0, 0, 0.71828, 0.599976, 0.611, nan ]

Clang results

Comparison Rule: BufferFloatEpsilon
Epsilon: 0.0008
Expected:
---
Name:            ExpectedOut
Format:          Float32
Stride:          16
Data:            [ nan, nan, 0, 0, 0, 0.5, nan, 0, 0.71828, 0.599976, 0.611, 
                   nan ]
OutputProps:
  Height:          0
  Width:           0
  Depth:           0
...
Got:
---
Name:            Out
Format:          Float32
Stride:          16
Data:            [ nan, 0, 0, 0, 0, 0.5, 0, 0, 0.71828, 0.599976, 0.611, 
                   nan ]
OutputProps:
  Height:          0
  Width:           0
  Depth:           0

DXC results

Expected:
---
Name:            ExpectedOut
Format:          Float32
Stride:          16
Data:            [ nan, nan, 0, 0, 0, 0.5, nan, 0, 0.71828, 0.599976, 0.611, 
                   nan ]
OutputProps:
  Height:          0
  Width:           0
  Depth:           0
...
Got:
---
Name:            Out
Format:          Float32
Stride:          16
Data:            [ nan, 0, 0, 0, 0, 0.5, 0, 0, 0.71828, 0.599976, 0.611, 
                   nan ]
OutputProps:
  Height:          0
  Width:           0
  Depth:           0
...
Full Hex 64bit representation of Expected Buffer Values:
[ nan, nan, 0x0p+0, 0x0p+0, 0x0p+0, 0x1p-1, nan, 0x0p+0, 0x1.6fc266p-1, 0x1.33300ep-1, 0x1.38d4fep-1, nan ]
Full Hex 64bit representation of Actual Buffer Values:
[ nan, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x1p-1, 0x0p+0, 0x0p+0, 0x1.6fc268p-1, 0x1.333p-1, 0x1.38d5p-1, nan ]