Fix saver's credit AGI boundary tiers

[saisharan0103]

Fixes #9151.

The saver's credit joint rate parameter stored IRS-published inclusive AGI maxima as single_amount thresholds. Since PolicyEngine treats those thresholds as lower bounds, AGI exactly equal to a published maximum moved into the next lower credit-rate tier.

This keeps the existing formula and filing-status adjustment logic, and shifts the nonzero joint thresholds to the first whole-dollar AGI above each published inclusive maximum for the years already represented in the parameter.

Also updates the nearby saver's credit YAML tests so exact published boundaries receive the higher inclusive-tier rate, adds the reported 2026 single-filer first-tier boundary regression case, and includes the repository-required fixed changelog fragment.

[MaxGhenis]

Merging under the cross-family dual-review protocol (Claude Fable 5 + GPT-5.6-sol, independent reviews, both merge-ok). Implements #9151's recommended next-whole-dollar representation of the IRS inclusive AGI maxima exactly.

Verified: every changed expectation recomputed from branch parameters; filing-status derivations (0.5/0.75) preserve whole-dollar boundary behavior; the changed scale's only formula consumer is savers_credit_person; 15/15 YAML cases pass. Nonblocking: fractional-AGI slivers between a published maximum and the +1 threshold retain the higher rate (documented whole-dollar convention per the issue); a maximum+1 fixture would pin the tier transition explicitly.