runtime_intrinsics.c: Correct max_double

[eschnett]

Closes #57119.

[adienes]

added tests would be prudent 🙂 (although I know you were not the pr original author)

[oscardssmith]

A test would be good, but this also seems to me like a really good reason for us to move away from runtime intrinsics. They're much more possible to break without people noticing.

[giordano]

I'm a bit confused by how this wasn't caught before. Regular uses of min/max work correctly, how are these intrinsics used?

[oscardssmith]

So there are 2 implementations of all of our intrinsics, the codegen implementation (in codegen.c or cgutils.c) which is the LLVM code that we generate when compiling the function, and the runtime.c implementation that gets called when the function is interpreted.

[giordano]

#57119 (comment) suggested tests and where to put them. For example, but not limited to:

julia> Core.Intrinsics.max_float(1.0, 2.0)
1.0

julia> Core.Intrinsics.max_float(1.0f0, 2.0f0)
2.0f0