Add Scotland and Wales private-rent calibration targets

[MaxGhenis]

Summary

This addresses #316 with a bounded target-side fix.

The root problem is not just sample size. Constituency calibration currently has no rent or tenure anchors at all, so Scotland and Wales can drift to implausibly low private-rent levels while still matching income, age, and UC targets.

This PR adds country-level Scotland and Wales anchors to the national objective used during constituency calibration:

• private-rented stock targets for Wales and Scotland
• linearised average private-rent constraints for Wales and Scotland
• tenure compute support for Wales/Scotland private-renter targets

Target values

2025 private-rented stock targets:

• Wales: 200,700
• Scotland: 357,706

2025 average private rent anchors:

• Wales: £795/month
• Scotland: £999/month

Why this shape

This does not claim to solve within-country differentiation like Edinburgh versus rural Scotland. We still do not have constituency-level rent targets.

It also deliberately avoids country rent-spend targets, because those would conflict with the current rough UK-wide rent total target. Instead, the PR uses a linear average-rent constraint:

• weighted private rent total
• minus target average rent × weighted private-renter count
• equals zero

That gives the optimizer the right signal without introducing a contradictory aggregate objective.

Sources

• Wales dwelling stock estimates at 31 March 2024
• Scotland stock by tenure 2023 workbook
• ONS private rents bulletin, May 2025

Testing

Passed locally:

• uvx ruff check policyengine_uk_data/targets/sources/devolved_housing.py policyengine_uk_data/targets/compute/households.py policyengine_uk_data/targets/compute/other.py policyengine_uk_data/targets/build_loss_matrix.py policyengine_uk_data/tests/test_devolved_housing_targets.py policyengine_uk_data/tests/test_target_registry.py
• uv run pytest -q policyengine_uk_data/tests/test_devolved_housing_targets.py policyengine_uk_data/tests/test_target_registry.py policyengine_uk_data/tests/test_housing_targets.py

[vahid-ahmadi]

Two items worth addressing:

1. households.py refactor is a no-op — should be removed or completed

   The compute_tenure refactor changes _TENURE_MAP values to (string, country) tuples and generalises the lookup, but the map still only contains tenure_england_* keys. No Wales/Scotland entries are added, and no source module provides Wales/Scotland tenure targets. This function is used by the general target registry, not the constituency calibration path — so it's not exercised by anything new in this PR. Should either be removed from the PR or completed with actual new entries.

2. Minor inconsistency: rent vs benunit_rent

   The existing local_authorities/loss.py:227-228 uses benunit_rent mapped to household level for its private rent targets. This PR uses rent (the household-level hhrent from FRS) directly. These could diverge for multi-benefit-unit households. Probably fine since the constituency path builds its own matrix, but worth a quick check that the same rent concept is intended.