docs(governance): state the boundary — keel enforces rulings, it does not issue them

[eaitbrahim]

Closes #280

Once contributors arrive from different madhāhib and jurisdictions they will disagree about
rulings, and in a religious context ungoverned disagreement fragments the project. The
architecture already answers the question — a classification is a human input
(keel assets attest --source --attested-by), screen_asset enforces it, attestation=None
is a rejection — so this PR makes the answer explicit, before any other Phase 6–8 document is
written against the wrong assumption.

What

• README's first screen (before the first ##, where a stranger decides what this is):
  the quotable sentence — keel is not a fatwa engine. It is an enforcement engine for a
  ruling you supply. — plus the two-operators-two-schools consequence.
• CONTRIBUTING.md: the same sentence verbatim, framed as PR scope:
  • a PR changing a default classification needs a cited source and is discussed —
    it applies one contributor's fiqh to every operator who upgrades;
  • a PR changing the mechanism is ordinary engineering;
  • disagreement routes locally — keel assets attest writes your ruling to your
    database — and is not merged upstream, so the project never becomes a fiqh court with a
    merge button.
• tests/test_governance.py (written first, red, then green): boundary sentence on the
  README first screen; same sentence verbatim in CONTRIBUTING (drift-proof); both PR kinds
  told apart with their bars; the local-not-upstream route named.

Acceptance from #280

• Short, quotable statement in the README's first screen
• Same statement in CONTRIBUTING.md, framed as PR scope
• Explicit: default-classification PRs need a source and discussion; mechanism PRs are
  ordinary engineering
• Stated disagreement route that does not require the project to adjudicate

Gates: pytest 2778 passed / 1 skipped, ruff check clean, mypy clean.