docs: lead the generative AI disclosure with human design authority

[shivamlalakiya]

What

Reorders the generative AI disclosure in paper.md, README.md, and
philanthropy/__init__.py so each one opens with what the author decided and
reviewed, then states the scope of the AI assistance within those constraints.

Why

JOSS requires the AI usage disclosure as a named section, and the January 2026
policy is explicit that a pro forma disclosure is a problem. The previous
ordering led with the tool and reached the human review gate last, which reads
as a tool-first disclosure even though the substance was already there.

The facts are unchanged in all three places:

• the scope is still package-wide, not per-module;
• no numeric split between authored and AI-assisted output is offered, because
  the author has not measured one;
• the disclosure still rests on the review gate (flake8, mypy, doctests, the
  suite against both coverage floors, the OS and Python matrix, the
  dependency-floor install, the packaging check, the conformance battery),
  not on a percentage.

Only the order and the emphasis moved, which keeps paper.md consistent with
README.md, AGENTS.md, and the public commit history.

Verification

• make ci: 1977 passed, 4 skipped, total coverage 97.77% against the 92.0% floor. Exit 0.
• make riskcov: risk-tier TOTAL 97%. Exit 0.
• draft-pdf will re-render paper.md on this PR; the paper stays within the
  1,750-word ceiling (no words added on net).

Credit

CHANGELOG.md entry added under ## [Unreleased] / ### Changed; author already
listed in CONTRIBUTORS.md.