Set Marriage Allowance take-up rate to HMRC outturn (#623)

[vahid-ahmadi]

Summary

• Issue Marriage allowance take-up rate #623: the Marriage Allowance take-up rate parameter was a placeholder 1.0 (100%). HMRC outturn shows ~2.1m claimants of ~4.2m eligible couples, roughly 50% take-up.
• Update takeup_rate.yaml to 0.5 with a House of Commons Library citation.
• Document in the parameter description that the load-bearing rate for microsimulation outputs lives in policyengine-uk-data and must be kept in sync.

Behavioural scope

In the baseline policyengine-uk model, the main marriage_allowance.py formula reads would_claim_marriage_allowance, a pre-generated 0/1 flag populated stochastically at FRS build time in policyengine-uk-data. For that baseline microsimulation path, the matching policyengine-uk-data parameter is the load-bearing one.

However, this parameter is still read directly by the CPS marriage-tax reform in policyengine_uk/reforms/cps/marriage_tax_reforms.py. So this PR is not strictly documentation-only across all code paths: it also changes take-up under that reform path from 100% to 50%, which appears directionally correct.

A parallel policyengine-uk-data PR is still required to move baseline microsimulation outputs. See PolicyEngine/policyengine-uk-data#369 and #623 for context.

Test plan

• make format clean
• Reviewer to confirm the 0.5 figure matches the team's preferred outturn anchor (HoC briefing cites HMRC)
• Reviewer to confirm the CPS reform path should also inherit the 0.5 take-up assumption

[vahid-ahmadi]

Review note: I checked the code paths behind this change. The parameter update itself looks coherent, but the original PR description was overstating the scope. In baseline policyengine-uk, marriage_allowance is driven by the pre-generated would_claim_marriage_allowance flag from policyengine-uk-data, so the dataset-side parameter is still the load-bearing one for baseline microsimulation. However, the CPS marriage-tax reform in policyengine_uk/reforms/cps/marriage_tax_reforms.py still reads allowances.marriage_allowance.takeup_rate directly, so this PR also changes behaviour there from 100% to 50% take-up. I updated the PR body to reflect that distinction.

[MaxGhenis]

Verified consistent with uk-data#369 (same 0.5 value, same effective date, same HoC SN00870 citation). uk-data#369 just merged. Approving.