fix: reject NaN values in FloatRange

[PewterZz]

Fixes #3347

FloatRange currently accepts NaN because NaN comparisons (NaN < min, NaN > max) always return False, causing the range check to pass. This bypasses validation entirely, including clamp=True mode.

The fix adds a math.isnan check in FloatRange.convert before returning the value. NaN is not a valid number for any range, so it should always be rejected.

Change summary:

• Override convert in FloatRange to check for NaN after calling super().convert()
• Fail with "nan is not a valid number" if detected
• 6 new test cases covering bounded, unbounded, clamped, and open-bound FloatRange configurations

All 1391 tests pass.

[davidism]

AI slop. I don't see why it matters if we accept all float values, and if certain values that are known not to compare don't compare.