WAS wealth imputation with the ISA split, plus regional property uprating

[juaristi22]

Workstream E5 of the UK-migration master epic #665; child of #145. Supersedes draft PR #181 (stranded on pre-Frame paths — salvage the WAS column mappings and citations, not the code). Closes #180.

Scope

• WAS round 8 household tab ingest (sha+size-pinned); cleaning: corporate_wealth composition, student_loan_balance = total_loans − total_loans_exc_slc, NI→WALES region remap (WAS does not sample Northern Ireland — a documented UK-specific divergence).
• Seeded weighted QRF (microcosm.fit.RegimeGatedQRF) for the 11 wealth outputs including num_vehicles (which the consumption stage consumes — this stage must precede E6). Formula-computed predictors (household_net_income, capital_income, council_tax) materialize through the declared rules-engine adapter.
• ISA split (UK: surface cash ISA and stocks & shares ISA holdings in the wealth imputation #180): cash_isa ← DVCISAVR8, stocks_and_shares_isa ← DVIISAVR8, with the investment-ISA component fold_into corporate_wealth for back-compatibility.
• student_loan_balance person-allocation waterfall (current repayers → reported borrowers → tertiary-qualified adults → current students → working-age → everyone) as a generic allocate_within_group_waterfall operation.
• Deterministic regional property uprating stage (pinned regional_land_values.csv; WAS compresses London ~41% low and the North East ~39% high).
• Optionally the WAS debt columns (policyengine-uk-data#426 mortgage_debt/consumer_debt) here rather than in the net-new-stages child.
• QRF model caching: the shared, metadata-keyed cache tooling ships default-off — builds refit per-run from seeds.

Acceptance

• The 11 outputs' weighted totals within stage tolerance vs the licensed input-mass register (deltas published SDC-safely).
• owned_land gets an explicit support/stability receipt (policyengine-uk-data#448 records its release-to-release instability) — expected to fail naive parity; signed with the evidence, likely a reviewed exclusion.
• Tail concentration recorded per output via the UK parity increment 4: port the weighted integrity gates (input-mass parity + QRF tail concentration) into the UK terminal battery (#578) #609 instrument; nonnegative_columns and support gates declared for the stage family.

UK-specificity note (required in the PR)

WAS column semantics and the NI remap are UK; QRF fitting, fold_into, and the waterfall allocation should be generic operations. List anything beyond that and why.

Depends on: E2/E3 (predictors); the E1 adapter.