floating point types have an incorrect implementation of Eq and Ord

[thestinger]

There are two solutions available:

• implement the total ordering for Eq and Ord, and offer the traditional comparison operators via intrinsics and the Float trait
• drop the implementations of Eq, Ord and traits inheriting from these like Num (alternate types could implement the total ordering)

Either way, it's an ugly mess. I no longer think a consensus can be reached to change the definition of the operators to using the totalOrder predicate due to both performance concerns and the desire to have the usual semantics so that's not part of the RFC.

[thestinger]

The pull request isn't going to land, so this can be closed as it was opened pre-emptively in order to refer to the issue in a FIXME.