_filter_outliers used lot.get("price_per_m2", 0), whose default only applies
when the key is absent; a present-but-None value (listings.price_per_m2 is
nullable) leaked into `low <= None <= high` → TypeError, crashing the estimate.
Make the bounds check None-safe: keep unpriced lots, still drop genuine priced
outliers. Behavior byte-identical for all real prices (incl. 0).
Add tests/test_estimator_pure_units.py (23 tests) covering _percentile,
_filter_outliers (incl. the None regression) and _compute_confidence
(thresholds, concentration downgrade, fallback notes).
Closes#580
