f32 sample/threshold domain is unconstrained and ErrorKind::InvalidInput is undocumented

Problem

Sample and threshold values are bare f32 with no documented domain constraints. The trait contracts are silent about:

• What NaN / inf mean as a sample or as a threshold argument.
• Whether Percentage is expected to be within 0.0..=100.0.
• What happens when threshold_low > threshold_high.

ErrorKind::InvalidInput exists ("The sensor was configured with invalid input") but no documentation anywhere tells an implementer when to return it, so in practice it is unreachable and generic code cannot rely on it.

Suggested fix

Pick one and be explicit:

1. Document implementer obligations on ThresholdSet / Hysteresis: e.g. "Implementations MUST return ErrorKind::InvalidInput if threshold is not finite, or if setting it would make low > high."
2. Validate at construction via newtype constructors (see the newtype issue) so invalid values are unrepresentable.

Option 2 is cleaner but breaking; option 1 is non-breaking and can land immediately.

Read the ThresholdSet and Hysteresis trait contracts and the ErrorKind::InvalidInput definition. Decide between documenting implementer obligations and validating through newtype constructors, then make the chosen behavior explicit for NaN or infinity, Percentage bounds, and reversed thresholds. Done means generic code can determine when InvalidInput applies.