Fix the Ord instance for FP and FloatingPoint

[lsrcz]

This pull request fixes the Ord instance for FP and FloatingPoint. The operator (>=) and (<=) will give True when comparing NaNs.

I originally thought that this comes from the underlying libBF library, as I've fixed a bug there yesterday. (GaloisInc/libBF-hs#32). However, after I switch to the fixed version, the bug is still there.

My theory for this bug in the sbv repo is that the instance derived by the compiler may look like this:

a <= b = not $ a > b

And the fix would be manually implementing the operators.

[LeventErkok]

I think we should we throw an error here if (eb0, sb0) /= (eb1, sb1), and if they're equal it should simply compare v0 and v1

My motivation is that if this gets called with a differing sized float, then that's really a type error. (i.e., it should never happen) Don't you think?

[LeventErkok]

Also, we should bump libBF dependency to require >= 0.6.8 I suppose.

[lsrcz]

Yeah that's a good idea. I implemented it like this just because this may be the original semantics, but fixed. Let me update it to throw errors.

For the libBF dependency, its (<) and (<=) are good, so we are able to implement a correct Ord instance even if the dependency is buggy.

[lsrcz]

I've updated it to throw errors on comparing different precisions.

[LeventErkok]

Great!

I'll make a stylistic change (I don't like otherwise, my pet peeve), and will bump the libBF dependency.