Correct the docs for WASM relaxed_simd's relaxed_min/relaxed_max operations

[valadaptive]

The current doc comments say that these behave like one of two operations:

• fmin/fmax, which return NaN if either operand is NaN, and treat -0.0 as less than 0.0 (this corresponds to IEEE754's minimum and maximum operations).
• fpmin/fpmax, which return the first operand if either operand is NaN, and also returns the first operand if both are ±0.0.

These are not the only possibilities, however! The specification says that it can also:

• Return the non-NaN operand if the other is NaN, and treat -0.0 as less than 0.0 (this corresponds to IEEE754's minimumNumber and maximumNumber operations).
• Behave like fpmin/fpmax but with the operands flipped.
• Possibly non-deterministically return ±0.0 if both operands are signed zero? The spec is very hard to parse here.

I've chosen to just link to the spec; it's easier than trying to explain the precise semantics.

[rustbot]

r? @Amanieu

rustbot has assigned @Amanieu.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer