Port the FRS derived layer: council tax, disability categories, education, legacy proxies, sector/SIC #679

Workstream E3 of the UK-migration master epic #665; child of #145.

Scope

Port the incumbent's FRS-derived layer as declarative stages:

• Council-tax imputation: (region, band, single-adult) cell means, ported post-uk-data#449 (Scottish Water netted out of CTANNUAL). Record per-nation deltas vs the incumbent and sign them — do not silently replicate the Scottish/Welsh overstatement (policyengine-uk-data#448); the durable devolved fix belongs with the scotgov band targets in WS-C/D. Prefer a generic impute_cell_means operation kind.
• Disability benefit categories from the 5 internal reported amounts (thresholded against parameter rates with the incumbent's tolerance; internal amounts stripped before any artifact is saved).
• Education: qualification maps, current-education derivation, UC qualifying-young-person flags, education-grant splits (the DSA maximum is a parameter read via the deferred-import pattern, not a simulation).
• Legacy-benefit proxies: the survey-condition logic ports directly (explicitly labeled proxies, not entitlement).
• Employment sector / SIC division (policyengine-uk-data#433 port), with the ONS public-sector target family noted for WS-C.

Acceptance

• Per-stage parity evidence for each declared output vs the frozen reference; council-tax per-nation deltas signed; disability category shares within tolerance.
• Declaration-lockstep test per stage; determinism per the WS-E harness.

UK-specificity note (required in the PR)

Council-tax band structure, FRS benefit codes, and QYP rules are genuinely UK; cell-mean imputation and threshold-categorization operations should be generic. List anything beyond that and why.

Depends on: E2.