Add more assertion helpers for floating-point tests

[valadaptive]

Right now, the tests for floating-point operations perform either a floating-point comparison or a bitwise comparison. There are issues with both of these approaches:

• Floating-point comparisons treat -0.0 == +0.0, and NaN != NaN.
• Bitwise comparisons care about NaN payloads, but LLVM does not mandate that NaN payloads be deterministic.

We should add a new assertion that treats -0.0 == +0.0, and all NaNs as equal.