Add experimental UK firm generator

[MaxGhenis]

Summary

• add an experimental UK firm generator under populace.build.uk_runtime
• add uk_firm_source_data_from_ledger_facts, which consumes Ledger/Arch consumer facts from the uk_firms target profile and reshapes them into the temporary firm-generator matrix contract
• consume the detailed Ledger firm target surface: ONS SIC-by-turnover, ONS SIC-by-employment, HMRC VAT-registered firms by SIC, and HMRC net VAT liability by SIC
• use Populace's shared calibration optimizer for firm weights
• document the hard target-source boundary: production Populace calibration targets are Ledger-owned, and raw administrative tables are only Ledger inputs or explicitly experimental migration-comparison inputs
• keep the processed-table reader only for paper-repository comparisons

Ledger dependency

• depends on Add UK firm calibration targets ledger#67 for the UK firm Ledger source packages and policyengine_ledger.target_profiles/uk_firms.json
• US fiscal-target build (0c38bc4): income tax regressed ~15% and SS dependents overshoot ~60% vs prior build #67 now covers the current paper target surface used by this adapter
• VAT metrics still use the paper generator's turnover/input heuristics; production activation should wait for Axiom-backed VAT metric execution

Tests

• uv run ruff check packages/populace-build/src/populace/build/uk_runtime/firm_generation.py packages/populace-build/tests/test_uk_firm_generation.py packages/populace-build/README.md
• uv run python -m pytest packages/populace-build/tests/test_uk_firm_generation.py
• cross-repo smoke: emitted 1,439 consumer facts from arch-data#67, rebuilt 88 ONS turnover rows, 88 ONS employment rows, 88 HMRC sector rows, and generated ~2.946M firm rows with a 4-iteration calibration

Review cycle

• addressed reviewer findings on VAT-registered target alignment, preserved employment calibration assignments, data_vintage target selection, and exact zero-turnover allocation
• final read-only review reported no actionable findings before the detailed Ledger-target follow-up