test(e2e): credit coverage for runs that span a source boundary

[hivecommons-hive]

Test Improvement

tests/tools/e2e-coverage-report.mjs — convertScript / the new
sourceBoundaryOffsets + splitRangeAtBoundaries helpers — plus a new
tests/e2e-coverage-source-boundary.test.mjs. No other file is touched.

The defect

normalizeRanges rewrites a V8 record into the coarsest partition it can: one
range per run of equal counts, however far that run reaches. In a bundle that
carries several src/** modules such a run routinely starts in one module and
ends in another.

v8-to-istanbul cannot place that range. offsetToOriginalRelative
(v8-to-istanbul lib/source.js) maps the start and the end back through the
source map and gives up when they disagree:

if (start.source !== end.source) {
  return {}
}

applyCoverage then returns before recording a branch, so neither module is
credited. convertScript builds attributedPaths from exactly the branches and
functions that were recorded, so the multi-source guard in collectE2ECoverage

if (converted.sourceFiles.length > 1 && !converted.attributedPaths.has(path)) continue;

goes on to skip every source of that script. Coverage a browser genuinely
executed never reaches the region tally — which is the shape #1236 describes.

The fix

Split each normalized range at the generated offsets where the source map
changes source, so every sub-range lies inside one original file. Each
non-final sub-range ends one offset short of the cut, because the first offset
at a boundary already belongs to the next source and would re-trigger the
start.source !== end.source bail; that offset is not lost, it opens the
following sub-range. A 1:1 bundle has no boundaries, so no range is split and
single-source scripts are byte-identical to before.

Evidence

Replayed deterministically against the e2e-coverage artifact 11609771627
(run 37914482229-1, head b00ff4e) named in #1236 — the artifact carries its
own scripts/ copies, so no browser is needed:

	regions	src region %	uncovered regions
b00ff4e as-is	447/466	95.92	19
with this diff	649/665	97.59	16

Line coverage is unchanged at 100.00% (2099/2099). 199 regions that the report
could not see before are now measured. Concretely,
src/components/MemberDirectory/DirectoryFreshness.js drops lines 30, 44 and 49
from uncoveredRegions (19 30 44 49 → 19), and
src/components/MemberDirectory/utils.js becomes measurable for the first time.

npm run test:unit: 2089/2089 pass. The three new tests are red without the
tests/tools/ change (2 fail) and green with it; the single-source test passes
both ways, which is the point of including it.

What this does not close

• src/components/CommunityPeople/index.js line 134 — the specific instrumented
  case [quality] coverage executed on e2e fixture pages is not credited: 19 src regions stay at zero across four runs #1236 opens with — is still reported uncovered after this change. The
  boundary defect is real and is the bulk of the loss, but it is not the whole
  of that one symptom.
• Step 4 of the issue (ratcheting --check-source-regions past 95 in
  .github/workflows/ci.yml) is not here. A contributor-tier token cannot push
  a diff touching .github/workflows/**; GitHub rejects it server-side. With
  this change the measured number is 97.59, so the threshold has room to move.

Related Issue

Refs #1236 (needs-human: src/components/CommunityPeople/index.js:134 is still uncovered after this fix, and ratcheting --check-source-regions in .github/workflows/ci.yml needs workflow-write permission this agent does not have)

Overlap with open held PRs

tests/tools/e2e-coverage-report.mjs is also touched by #1233 (parseArgs,
main) and #1211 (the constants block near the top). This diff is confined to
convertScript and two new helpers directly above it, so it shares no hunk with
either.

---

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.

[mrbobbytables]

Heads-up: PRs1211 1233 also edit tests/tools/e2e-coverage-report.mjs. Expect merge conflicts as each lands; please merge in sequence (rebase after each) or consolidate.