test(e2e): cover the istanbul readers in e2e-coverage-report.mjs

[hivecommons-hive]

Test Improvement

Refs #1210 — the coverage work is done here, but the issue also asks for the
per-file region floor to be ratcheted and that single line cannot be written
yet (see Scope / overlap below), so the issue stays open to track it.

This takes #1210's option 1 (export the readers) for the part that option
covers, and its option 2 (record the residual) for the part it does not.

tests/tools/e2e-coverage-report.mjs exports only sourcePathFromReference,
collectE2ECoverage, renderE2ECoverageReport, writeE2ECoverageReport and
main. The guards inside its four module-private istanbul readers —
getLineCoverage, getRegionCoverage, isPhantomRegion and
isContradictedRegion — defend against malformed istanbul objects, and
istanbul objects are built inside the module by v8-to-istanbul from the V8
payload. Reaching a guard from the public surface requires that library to emit
a shape it does not emit, which is why no probe driven through
collectE2ECoverage in #1208/#1209 could close them.

Exported the four readers for tests (tests/tools/e2e-coverage-report.mjs)
and added tests/e2e-coverage-report-readers.test.mjs — 19 tests driving the
guards directly with hand-built istanbul objects:

• getLineCoverage — absent s; statements whose map entry is missing, has no
  start, or has a non-integer line; the highest-count-wins fold on one line.
• getRegionCoverage — absent b; a branch with no branchMap entry and one
  with no locations; null/empty/non-integer-line locations; the
  end/column defaults; a branch with no counts array and one with a short
  one.
• isPhantomRegion — executed region, single-line zero, inverted span, the
  statement-free line skipped mid-span, and the unexecuted line that keeps the
  zero.
• isContradictedRegion — executed region, no script recorded the zero, a
  witness with no executable line in the span, a witness that left a span line
  unexecuted, a witness carrying a zero over the span, the same-script witness
  that does contradict, and the other-script witness that is correctly ignored.

Recorded the residual that exporting cannot reach, as an in-file note at
the top of the module naming each entry by function and expression rather than
by line so it does not go stale: normalizeSourceMap's
map.resolvedSources ?? … (@jridgewell/trace-mapping internals),
convertScript's converter.branches ?? {} / converter.functions ?? {} and
the .map(([path]) => path) arm behind them, collectE2ECoverage's
converted.coverage ?? {} and if (lineCoverage.size === 0) continue, and
main's error.message ?? String(error) / error.stack ?? error.

Measured

npm run test:unit:coverage, node v26.10.0, TZ=UTC, run locally at
593b9fe (this branch's base):

	lines	regions
before	100.00	93.65
after	100.00	96.28

All-files regions 95.20 → 95.34. Full unit suite: 2105 tests, 0 failures.
npm run check and npm run test:unit:coverage:check both exit 0; Prettier
clean.

Scope / overlap

• Touches only tests/tools/e2e-coverage-report.mjs and one new test file.
  No open PR touches either. test(e2e): cover the e2e coverage reporter's partial-capture fallbacks #1209 adds
  tests/e2e-coverage-partial-capture.test.mjs against the public surface;
  the regions it closes (295-296 671 681 683 750 at this tree) are disjoint
  from the reader guards here and are deliberately not claimed by the in-file
  note.
• Deliberately not touching the per-file floor. [quality] record or expose the 18 unreachable guard regions in e2e-coverage-report.mjs #1210 asks for
  --check-harness-file-regions to be ratcheted, but that flag does not exist
  on main — it is added by open PR test(coverage): give the tests/tools/ harness a floor of its own #1207 in package.json. Ratcheting it
  here would collide with that PR over the same line. It should be raised to at
  least 96 in a follow-up once test(coverage): give the tests/tools/ harness a floor of its own #1207 lands.

---

Filed by quality agent (hold-gated mode). Human review required.

[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.

[mrbobbytables]

Heads-up: PRs 1233 1240 also edit tests/tools/e2e-coverage-report.mjs. Expect merge conflicts as each lands; please merge in sequence (rebase after each) or consolidate.