Add LA-level household land value calibration targets

[vahid-ahmadi]

Summary

Add ons/household_land_value/{la_code} calibration targets for all 360 UK local authorities, generalising the regional method introduced in #315.

Motivation

targets/sources/mhclg_regional_land.py distributes the ONS National Balance Sheet household land total across the 11 GB regions using dwellings × avg_house_price. There is no equivalent at local-authority level, so:

• The LA reweighting loss (datasets/local_areas/local_authorities/loss.py) currently has no property- or land-value anchor — its targets cover age, income, UC, tenure and rent only. As a result, per-LA avg household land values from LA-reweighted microdata are compressed at the top end and noisy at the bottom (e.g. Kensington and Chelsea averaging ~£240k, East Renfrewshire below £50k, neither of which matches the UK HPI ordering).
• Downstream analyses that want an LA map of household land value — e.g. a UK LVT blog post using this dataset, or constituency-level reform impacts — currently have to bolt on ad-hoc methods.

Proposed approach

Direct generalisation of the regional formula:

share_la  = (households_la × avg_house_price_la) / Σ_LAs
target_la = share_la × HOUSEHOLD_LAND_VALUES[year]

Data sources (all already used elsewhere in the repo):

• HM Land Registry UK HPI — average house price by LA, monthly. Same source used for regional_land_values.csv, just at LA granularity.
• LA household counts — derived from the existing storage/local_authority_weights.h5 matrix (sum of each LA's weight row), which keeps household-count semantics consistent with the rest of the LA calibration.
• National anchor — HOUSEHOLD_LAND_VALUES in targets/sources/_land.py (ONS National Balance Sheet 2025).

The new targets/sources/la_land.py module exports 360 Target objects with geographic_level=GeographicLevel.LOCAL_AUTHORITY, following the mhclg_regional_land.py layout.

Acceptance criteria

• 360 targets named ons/household_land_value/{code} registered via get_all_targets().
• LA targets sum to the national HOUSEHOLD_LAND_VALUES[year] for each year in the series (within 1e-6).
• Avg household land value in Kensington and Chelsea > 3× Blackpool.
• London boroughs dominate the top quintile of LAs by avg household land value.
• CSV covers all four UK countries (E/W/S/NI) and every code in local_authorities_2021.csv.
• Existing test_regional_land_value_targets.py stays green (filter by GeographicLevel.REGION where it previously filtered by name prefix only).

Out of scope for this issue

Wiring these targets into datasets/local_areas/local_authorities/loss.py so the LA reweighting actually calibrates against them. That should be a follow-up PR once this one lands.

References

• Regional property wealth calibration: WAS-imputed main_residence_value doesn't match UK HPI by region #314 — original national → regional land value work
• Add regional household land value calibration targets #315 — regional calibration targets (the pattern this issue generalises)
• UK HPI: https://www.gov.uk/government/statistical-data-sets/uk-house-price-index-data-downloads-december-2025
• ONS National Balance Sheet 2025: https://www.ons.gov.uk/economy/nationalaccounts/uksectoraccounts/bulletins/nationalbalancesheet/2025