docs: state the mechanism directly in the entry-point READMEs

[kmadan]

What this changes

Rewrites the introductory sections of the three entry-point READMEs so they read
as reference documentation rather than narration, and corrects several imprecise
descriptions. Documentation only; no .rego file is touched.

Type

• New policy (specify framework path)
• New framework directory
• Policy update (semver MINOR — backward compatible)
• Policy update (semver MAJOR — breaking; new v2/ directory)
• Bug fix
• Tooling / CI / docs

Detail

README.md, examples/model-switch/README.md and docs/coverage/README.md
mixed two registers. Their reference sections read as documentation; their
introductory sections staged behaviour as a sequence of events, used fragments
for emphasis, and put unnamed people in place of the component doing the work.
Those sections now state the condition, the mechanism and the result, with the
policy, the score or the threshold as the grammatical subject.

Four descriptions were also imprecise and are corrected:

• The model-switch caption described a live model swap. The example compares
  checked-in baseline and candidate toxicity metrics with the prompts, the
  scoring method and the policy held fixed.
• "Why GOPAL" characterised the alternatives by naming vendor categories. It now
  states what generic policy-as-code and proprietary governance platforms each
  supply, and where an open AI-regulatory policy library differs. A Portable
  point is added, covering opa eval, Conftest, an existing OPA server and
  AICertify.
• The declared/measured distinction now states why a supplied number remains a
  declaration unless the evaluator evidence behind it is supplied as well.
• The EU AI Act figure is stated as 185 input-field reads rather than 185
  fields, which is what the count measures; unique declared fields are fewer.

An earlier correction rides along in 1685c11: 4321e7a moved the EU split to
170 declarations and 15 measurements but left the following sentence reading 14.

Every claim, figure, link and heading target is preserved. Verified against the
tree: 29 EU AI Act policy files, and the 170/15/185 split unchanged from main.

Checklist

• Every new .rego has a sibling *_test.rego — n/a, no .rego changed
• Metadata block present on every new policy — n/a, no new policy
• Framework-level README.md lists the new policy — n/a, no new policy
• opa check --ignore custom/ . passes locally
• regal lint --ignore-files custom/ . passes locally
• Not breaking; no v2/ directory required

Also run locally, all passing: check-anchors.sh, check-version-refs.sh,
check-test-coverage.sh, check-deprecations.sh, generate-coverage.sh --check,
model-card-coverage.sh --check.

Source(s)

No regulation is encoded or reinterpreted by this PR.

Notes for reviewers

These two commits were pushed to chore/zenodo-2-0-0-doi after #90 had already
merged, so they were stranded on a closed PR's head branch and never reached
main. This PR carries them across unchanged; nothing is rebased or squashed.

Four framework READMEs (examples/github-actions/, industry_specific/,
international/, international/uk/) carry one-line wording changes from
1685c11 and are included in the diff for that reason.

Reviewing 86a986e and 1685c11 separately is easier than reading the combined
diff, since the second commit rewrites paragraphs the first had already touched.