test: let data overlays patch a root-level JSON array with setWhere

[hivecommons-hive]

Test Improvement

tests/tools/e2e-data-fixtures.cjs addressed data through dotted paths walked
down from a root object, so parentOf() rejected an array at every step. That
left the two data files whose root is a JSON array — data/architectures/catalog.json
and data/projects-born.json — outside the overlay mechanism entirely.

tests/tools/e2e-data-fixtures.cjs — new setWhere overlay key. It selects
one record of a top-level array by <field>=<value> and patches its fields,
keeping the fail-loud property the dotted paths have: a selector matching no
record, matching more than one, or naming a field the record does not carry is
an error rather than a silent no-op. It is keyed on a field rather than a
positional index because both files are regenerated (npm run import:architectures,
npm run collect:projects-born) and entry order is not stable.

tests/e2e/fixtures/data/architectures/catalog.json — new overlay emptying
allianz's industries, so ArchitectureCard renders the eyebrow fallback at
src/components/ReferenceArchitectures/index.js:66. This is an additive case,
not a swap: the other seven entries keep their industries, so both arms of the
|| render on the same page in the same build. No variant build is needed.

tests/e2e/data-fixtures.spec.js — one spec driving the fallback card and
the populated card on the same page, so the assertion is evidence the branch
switched rather than that every eyebrow went generic.

tests/e2e-data-fixtures.test.mjs — pins setWhere's patch semantics and
each of its six rejections; pins that both root-array data files are now
addressable against the real files; adds the catalog overlay to the
"every committed overlay still applies" case, both directions.
tests/e2e-data-fixtures.test.mjs:85-91 ("setting a leaf whose parent is an
array is rejected") is unchanged — leaf-under-array addressing stays rejected.

Verification

• npm run test:unit — 1975 passed, 0 failed at 7ab301e + this diff.
• npm run build:e2e:coverage — succeeds. The built coverage site server-renders
  the fallback: build/architectures/index.html contains
  <a class=card_D3WH href=/architectures/allianz>…<p class=eyebrow_MUlS>Reference architecture,
  while the other cards still render their joined industries.
• npx prettier --check tests/ — clean.
• The Playwright spec itself could not be executed here: this sandbox has no
  libglib-2.0.so.0, so chrome-headless-shell exits 127 before any navigation,
  and all seven pre-existing specs in that file fail identically. That is
  environmental, not diff-local; CI's End-to-end coverage job is the first run
  of the new case.

Scope

Only tests/**. No production code, no workflow file. --check-source-regions
needs no change: this adds a covered region without adding a build.

Disjoint from the open hold-gated PRs — #1090 is scripts/lib/svg-active-content.mjs,
#1104 is tests/e2e/fixtures/data-variants/members.json plus its own spec file.
Neither touches tests/tools/e2e-data-fixtures.cjs, tests/e2e-data-fixtures.test.mjs,
tests/e2e/data-fixtures.spec.js or tests/e2e/fixtures/data/architectures/.
Also disjoint from the tests/tools/e2e-coverage-report.mjs region-attribution
cluster (#1066, #1079, #1097).

Related Issue

Closes #1086

---

Human review required.

— hive: agent=quality backend=copilot model=claude-opus-5 copilot=1.0.88

[hivecommons-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "quality" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will keep the hold label until a human removes it. Operators can make a deliberate one-off release during an ACMM level change with release_level_holds=true, but level changes never release this hold automatically.