test: cover MetricsDashboard's four empty-collection fallbacks end to end

[hivecommons-hive]

Test Improvement

src/components/MetricsDashboard/index.js maps over four document-level
collections of data/metrics.json through a fallback that supplies an empty
one, and the e2e report lists all four as uncovered:

line	expression	collection
65	(lifecycle?.cards || []).map(...)	lifecycle cards
85	(lifecycle?.omitted || []).map(...)	lifecycle "what is not yet measurable" items
124	Object.entries(metricsData.series || {})	line charts
174	Object.entries(metricsData.breakdowns || {})	bar charts

The checked-in document always carries four cards, three omitted indicators,
two series and two breakdowns, so none of the fallbacks is reachable in the
browser. They are ordinary shapes rather than defensive dead code:
data/metrics.json is regenerated by npm run collect:metrics, and a
collector run that cannot reach one of its sources emits the document without
that collection.

What this changes

• tests/e2e/fixtures/data-variants/metrics.json — the existing variant
  overlay additionally clears referenceArchitectureLifecycle.cards,
  referenceArchitectureLifecycle.omitted, series and breakdowns.
• tests/e2e/metrics-empty-collections-variant.spec.js (new) — pairs
  /metrics with /e2e-coverage-variant/metrics. The real route must render
  every collection; the variant route must render none of them while the
  lifecycle section, its heading, both trend labels and the Last updated
  line still render. Pairing the two is what makes this evidence that the arms
  switched rather than that one page happens to look a certain way.

Emptying the collections in the ordinary coverage build would trade every
rendered card, disclosure item, line chart and bar chart for the fallback
rather than add a case, which is why they belong to the variant build —
the same reason AwardsTimeline's provenance paragraph does.

Deliberately out of scope

referenceArchitectureLifecycle.trends (line 76) is left alone. The same
overlay appends a point to trends.submissions.values for
tests/e2e/metrics-sparkline-variant.spec.js, and applyOverlay in
tests/tools/e2e-data-fixtures.cjs applies set before append, so clearing
trends would make that append throw and fail the build. Reaching line 76
needs a second variant build, not a second entry in this overlay. The fixture
and the spec both say so in place.

Verification

• npm run build:e2e:coverage at 7ab301e with this change applied, then both
  rendered documents read directly:

  	build/metrics/index.html	build/e2e-coverage-variant/metrics/index.html
  lifecycleCard elements	4	0
  <p> inside the lifecycle <details>	3	0
  labelled section per series id	2	0
  labelled section per breakdowns id	2	0
  lifecycle section, heading, both trend labels, Last updated	present	present

• npm run test:unit — 1966 passed, 0 failed (the overlay is validated by
  tests/e2e-data-fixtures.test.mjs, which loads every committed overlay).

• npx prettier --check and npx eslint clean on both files.

• The Playwright run itself could not be executed in this agent's
  environment: chrome-headless-shell fails to start with
  error while loading shared libraries: libglib-2.0.so.0, and the sandbox has
  no root to install it. Every assertion in the new spec is instead verified
  above against the HTML the two builds actually emit; the browser run is left
  to the End-to-end coverage job on this PR.

Coordination

Touches only tests/e2e/fixtures/data-variants/metrics.json and the new
tests/e2e/metrics-empty-collections-variant.spec.js. It does not edit
tests/e2e-data-fixtures.test.mjs (#1113, #1110),
tests/e2e/data-fixtures.spec.js or tests/tools/e2e-data-fixtures.cjs
(#1110), or tests/e2e-coverage-report.test.mjs (#1114). The cases list in
tests/e2e-data-fixtures.test.mjs is hardcoded and not exhaustive, so a new
set entry needs no edit there. No workflow file is involved: the gate
--check-source-regions 80 only rises.

This PR does not promise the four regions leave the report's uncovered list.
tests/tools/e2e-coverage-report.mjs keys a region on its exact original
coordinates and the two builds' generated code does not always map back to the
same ones — that is the region-attribution problem tracked in #1066 and #1079.
Where it folds, these fold with no further change here.

Related Issue

Closes #1116

---

Filed by quality agent (hold-gated mode). 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.