AR low-income tax tables: 2021-2024 columns keyed $1 below the published $X,X01-$X,X00 rows (2025 column is correct)

[hua7450]

Found during the review of #9025 (backward-leak audit series: #9008, #9022, #9024).

Problem

The Arkansas low-income tax tables publish rows as FROM $X,X01 TO $X,X00 inclusive (verified verbatim in the 2021 instructions, 2024 instructions, and 2025 TaxTables_FI):

2021: $31,901  $32,000  $921
      $32,001  $32,100  $933
      $32,101  $32,200  $945
      *Above $32,200 use Standard or Itemized Deductions and Regular Income Tax Table
2024: $34,901  $35,000  $754
      $35,001  $35,100  $765
      *Above $35,100 ...

In parameters/gov/states/ar/tax/income/rates/low_income_tax_tables/joint/two_or_more_dependents.yaml, the 2025 column keys thresholds at the row lower bound (X_X01) — correct for the scale's at-or-above semantics. But the 2021–2024 columns key at X_X00 — $1 low on every boundary (in the 2024 column, 63 of 64 finite thresholds end in 00; in the 2025 column, 64 of 66 end in 01).

Effect

At exact-$100 incomes the wrong row applies, and at each year's cutoff the taxpayer is kicked to the regular table $1 early:

Year	Income	PE returns	Published
2024	exactly $35,000	$765	$754
2024	exactly $35,100	regular table (~$1,100)	$765
2021	exactly $32,200	regular table	$945

Scope

Verified in detail for joint/two_or_more_dependents.yaml; the same X_X00 keying pattern almost certainly affects the pre-2025 columns of all eight files under low_income_tax_tables/ (single, HoH, joint, surviving-spouse × dependent counts). The fix is mechanical (+1 on every pre-2025 threshold) but each file's columns should be re-verified against the year's published table during the shift.

Discovered while reviewing #9025 — the leak fix there is unaffected (its .inf scoping is threshold-value-agnostic).