Draft: Fix #156: Handle NaNs in comparisons correctly.

There are several parts here.

• Add vops to handle the comparisons <= and >= for both 32-bit signed and unsigned words and for single-float and double-float.
• Add two-arg->= and two-arg-<=
• Add generic->= and generic-<=
• Modify multi-compare so that <= isn't converted to >. Likewise for >=.

This is only implemented for x86 but probably should be done for other architectures.