feat: add student loan plan uprating to economic assumptions

[nwoodruff-co]

Summary

• Refactored uprate_student_loan_plans() in economic_assumptions.py to be idempotent — each year's output depends only on base year data and year parameter, not accumulated state from previous years
• Re-labels existing loan holders based on cohort start year: Plan 1 for pre-2012, Plan 2 for 2012–2022, Plan 5 for 2023+ in England
• Extends Plan 1/2 cohorts to cover age bands beyond base year's coverage:
  • Plan 2: extends from ages 21-29 (base year) to 21-35 by 2030
  • Plan 1: extends from ages 32-40 (base year) to 37-46 by 2030
• Uses highest_education == TERTIARY as signal for graduate status when adding new cohort members
• Applies flat 40% take-up rate (derived from SLC forecasts ~4M above threshold vs ~8-10M UK graduates)
• Writes off Plan 1 loans after 29 years

Changes from v1

• Fixed cumulative sampling bug: Previous version did independent stochastic draws each year, inflating counts over time
• Added Plan 1/2 cohort extensions: Now assigns loans to tertiary-educated NONE people in extended age bands
• Simplified take-up: Replaced unsourced income-decile table with flat 40% rate
• Removed test script: Cleaned up after validation

Validation results

Age distributions by year:

Year	Plan 1 ages	Plan 2 ages	Plan 5 ages
2025	32-46	21-31	19-20
2030	37-46	21-36	19-25

Totals (weighted):

Year	Plan 1	Plan 2	Plan 5
2025	1.62M	2.65M	0.02M
2030	0.71M	2.96M	0.90M

Plan 2 totals remain below SLC targets (~4-5M) due to FRS under-coverage of graduate cohort — reweighting tracked as follow-on work in policyengine-uk-data.

Test plan

• Age distributions extend correctly over time
• Plan 5 grows from scratch as new cohorts enter
• Plan 1 shrinks as loans are written off
• CI passes