Avoid `fmod` become ambiguous once CCCL specializes it for extended floating point types

[miscco]

I have a PR open that properly implements fmod and modf for all supported floating point types.

However, that creates an ambigouity with the fmod definition in scalar_ops, because the MatX half wrapper is convertible to both float and the half type.

Avoid this by specializing that function as we do for exp