manifest.py: numeric_threshold/age_interval_threshold with no 'threshold' field silently classifies zero rows

[yakew7]

Where: faircode/manifest.py's ProtectedAttribute dataclass - threshold: float | None = None has no required-ness validation for the numeric_threshold/age_interval_threshold types that depend on it.

Repro:

>>> from faircode.manifest import ProtectedAttribute
>>> import pandas as pd
>>> pa = ProtectedAttribute(name="age", type="numeric_threshold", column="age")  # no threshold
>>> pa.disadvantaged_mask(pd.DataFrame({"age": [20, 30, 40]}))
(0    False
 1    False
 2    False
 dtype: bool, 0    True
 1    True
 2    True
 dtype: bool)

numeric < None evaluates to all-False in pandas rather than raising, so a manifest with a missing threshold field silently produces zero disadvantaged rows for that attribute, and known incorrectly reports every row as classifiable.

Why it matters: this doesn't crash - it fails several steps downstream, where compute_metrics sees an all-False disadvantaged mask and reports "insufficient_data". A contributor debugging a confusing "insufficient data" result would have to trace back through several layers to discover the actual problem is a missing threshold field in their audit.yaml, not genuinely sparse data.

Suggested fix: raise a clear error (e.g. in ProtectedAttribute.__post_init__ or wherever manifests are loaded from YAML) when type is numeric_threshold/age_interval_threshold and threshold is None, naming the attribute and the missing field - mirroring the existing "need disadvantaged_values or advantaged_values" check already present for the categorical type.