Aggregate impact: record which small-cell condition withheld a reform-year #2654

The aggregate-impact tool (policyengine_uk/utils/aggregate_impact/measure.py, reform_withheld, added with #2335) withholds a reform's costing and target change in a year where any of three conditions holds:

1. the reform changes a record of a component suppressed in the baseline;
2. it leaves another component with 1 to 9 records;
3. it changes gov_balance in 1 to 9 households, or its target variable in 1 to 9 records.

The report records only that a year was withheld, not which condition fired.

Why this matters: the qualification gate (policyengine.py's reform_responsiveness) can't recompute the tool's verdict for a withheld year, so it returns needs_review and someone has to read the code to resolve it.

Example: in the reform-responsiveness.json receipt for Microcosm cut microcosm-uk-2024-25-national-20261023T075039Z-a133f37a (pe-uk 2.147.2, tool at 1191ac5a), reform_targets.additional_rate_plus_3pp is withheld in 2027 to 2030 and published in 2025, 2026 and 2031 (+£3.9bn to +£5.5bn). The baseline suppresses the same two heads, spending.disabled_students_allowance and spending.statutory_paternity_pay, in all seven years. So the receipt alone can't say why only 2027 to 2030 are withheld. The Budget coordinator accepted the gate for that cut (PolicyEngine/policyengine.py#715, comment). The tool's own responsive:additional_rate_plus_3pp check, computed before withholding, passes in every year.

Ask: record, for each withheld reform-year, which condition fired (for example withheld_because: suppressed_component_changed | small_component_left | small_target_change), without publishing any small-cell value or count. Check first that the condition label can't itself disclose a small cell, for example combined with the published years. Add a test.

Non-blocking: the release decision above doesn't depend on it.