[libclc] [libclc] Fix int<->float conversion builtins (PR #126905)

[Fraser Cormack via cfe-commits]

frasercrmck wrote:

CC @rjodinchr. I remember you saying you're OOO so no pressure, but perhaps you could verify this commit with clspv?

For clarity, the remaining failures I see are, e.g.,

```
Error for vector size 1 found at 0x00000001:  *-0x0p+0 vs 0x0p+0
Input value: -0x1.000002p-25  (convert_half_rtp( float ))
         *** convert_halfn_rtp( floatn ) FAILED **
Building convert_half_rtz( float ) test
Testing... ......
Error for vector size 1 found at 0x00000001:  *-0x0p+0 vs 0x0p+0
Input value: -0x1.000002p-25  (convert_half_rtz( float ))
         *** convert_halfn_rtz( floatn ) FAILED **
```

Looks like a problem with rounding to near zero to me. Does anyone else see this?

https://github.com/llvm/llvm-project/pull/126905