fix(coverage): fold e2e regions whose spans cross across artifacts

[hivecommons-hive]

What this changes

tests/tools/e2e-coverage-report.mjs now drops a zero-count region that a
covered region crosses: one that begins strictly inside the zero span and
ends strictly after it.

Two branch spans of one source file either nest or stay disjoint, because both
are AST extents. A crossing pair is not a shape the language can produce, so it
can only arise when two artifacts of the same chunk map one original branch to
different spans — which is exactly what #1079 measured:

file	zero region	covered twin
src/components/MemberDirectory/MemberProfile.js	10:40-10:48	10:44-15:19
src/components/MemberDirectory/MemberProfile.js	15:19-15:80	15:70-38:15
src/components/MemberDirectory/MemberCard.js	75:19-75:40	75:33-81:1

In all three the start columns differ, so the ordinal key proposed in #1066
keeps them apart, and neither span contains the other, so #1051's multi-line
fold leaves them standing.

getRegionCoverage() now keeps the start and end columns it already encoded
into the region key, which is what makes the comparison possible at all.

Why crossing and not containment

#1079's first recommendation was containment. The guard merged in #1114
established that containment is the wrong rule: a covered region that encloses
a zero one is the ordinary shape of a branch arm inside an executed block, and
GroupLinkStatus's checkedAt guard and RadarReports' empty-corpus arms are
enclosed exactly that way. Folding on enclosure would have erased real gaps that
#1094 and #1097 are writing tests for.

Crossing is strictly narrower. The test added in #1114 — "a single-line
zero-count region survives a covered region that encloses it" — passes
unchanged, including the same-start-column case it pins.

Tests

tests/e2e-coverage-report.test.mjs gains "a zero-count region crossed by a
covered one is treated as source-map drift", sitting directly beneath #1114's
guard so the two shapes are read together. Its source map carries mappings at
columns 0, 3 and 6 of line 1 so that a crossing is representable at all, and the
zero region 1:0-1:6 is crossed by the covered 1:3-3:12. The drifted half is
dropped from numerator and denominator, so no file is credited for a region
nobody observed.

• npm run test:unit — 2026 passed, 0 failed
• npm run test:unit:coverage:check — passes; tests/tools/e2e-coverage-report.mjs
  at 100.00% lines / 93.65% regions, with every added line covered
• npx prettier --check on both touched files — clean

Related Issue

Refs #1079 — the last completion criterion, re-deriving --check-source-regions
from the corrected number, is deliberately left open. That floor lives at
.github/workflows/ci.yml:257 (asserted by tests/workflow-scripts.test.mjs:355),
and a diff under .github/workflows/ is rejected server-side for this lane's
contributor-tier app token. Refs #1079 (needs-human: raising the e2e region
floor requires editing .github/workflows/ci.yml, which this lane cannot push).

Leaving the floor at 91 is safe in the meantime: it is a minimum, and this change
only raises the measured percentage.

---

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.