engine(ranker): opportunity-metadata.ts's blanket `v8 ignore` blocks hide untested scoring branches

[JSONbored]

⚠️ Definition of Done: this issue must be completed in full, in a single PR. Do not split this
work across multiple PRs, and do not defer any Deliverable below to a follow-up issue. A PR that
satisfies only some of the Deliverables, stubs a required test, or leaves a checkbox
partially-done does NOT resolve this issue and will be closed.

Context

packages/loopover-engine/src/opportunity-metadata.ts computes the five ranker inputs
(potential, feasibility, laneFit, freshness, dupRisk) that decide which issue the miner works
on next. The file is inside vitest.config.ts's coverage.include and carries the dedicated engine
Codecov flag, yet 18 v8 ignore directives suppress branch counting across nearly all of its decision
logic:

• computeMetadataPotential (lines 109-123) is wrapped entirely in a v8 ignore start/stop pair,
  and each of its four branches additionally carries its own v8 ignore next. It is a pure function of
  { labels } returning a number.
• computeMetadataFeasibility's non-finite-clock guard (line 130) and its whole title-length tier
  block (lines 135-141) are suppressed.
• titlesOverlap (lines 146-158) is suppressed, despite being deliberately re-exported for testing via
  opportunityMetadataInternals (lines 160-167).
• computeMetadataDupRisk's five branches (lines 180-192) are each suppressed.
• Six directives are bare — /* v8 ignore next */ with no rationale at all (lines 120, 204, 206,
  215, 219, 236) — against a file where every other directive carries a -- reason explanation. The
  repo convention is clearly to justify each one.

The rationales that do exist are self-refuting. Line 109 says "Metadata heuristics are exercised
end-to-end in test/unit/miner-opportunity-ranker.test.ts" and line 113 says "exercised in ranker
tests". If a line is genuinely exercised, v8 ignore on it is redundant — the coverage counter would
already show it hit. The directive only changes anything when the line is not hit. So each of these
is either dead weight or is concealing an untested branch, and today there is no way to tell which.

The concrete risk: these branches are the scoring rules. computeMetadataPotential's
labels.includes("refactor") bonus (line 120, bare-suppressed), computeMetadataDupRisk's
same-repo-peer filter (lines 184-188, all suppressed), and titlesOverlap's shorter.length >= 12
threshold could each be inverted or deleted today without a single test failing and without the
coverage gate noticing.

Requirements

• Every v8 ignore directive in packages/loopover-engine/src/opportunity-metadata.ts MUST be
  deleted — all 18, including the start/stop pairs and the six bare ones. The single exception is
  the opportunityMetadataInternals export block (lines 160-167), whose "Test-only export surface for
  branch coverage" rationale is legitimate and may keep its start/stop pair.
• No production logic in this file may change. The scoring formulas, thresholds, label lists, and
  return values MUST be byte-for-byte identical after the change.
• Direct unit tests MUST be added for each function whose suppressions are removed, exercising every
  branch that was suppressed (enumerated in the Deliverables below). These must call the exported
  functions (and opportunityMetadataInternals members) directly, not only through
  rankMetadataOpportunities.
• The tests MUST live in packages/loopover-engine/test/opportunity-metadata.test.ts (a new file), so
  the engine Codecov flag's own node:test suite covers them.

⚠️ Required pattern: follow the direct-unit-test style of
packages/loopover-engine/test/opportunity-ranker.test.ts and
packages/loopover-engine/test/opportunity-competition.test.ts — node:test + node:assert/strict,
importing from ../dist/index.js like every other suite in that directory. What does NOT satisfy
this issue: replacing a v8 ignore with a c8 ignore or an istanbul ignore; moving the
suppressions into vitest.config.ts's coverage.exclude or codecov.yml's ignore; deleting the
directives without adding tests (the patch gate would fail, but the branches would still be
untested); adding one end-to-end rankMetadataOpportunities test and calling the branches covered;
or changing any threshold, weight, or label list.

Deliverables

• packages/loopover-engine/src/opportunity-metadata.ts contains exactly one v8 ignore pair (the
  opportunityMetadataInternals export block) and no other v8 ignore directive.
• git diff shows no change to any numeric literal, label array, comparison operator, or return
  expression in that file.
• packages/loopover-engine/test/opportunity-metadata.test.ts exists and covers
  computeMetadataPotential for: a negative label present (returns 0), a positive label present,
  "bug" present, "refactor" present, "bug" + "refactor" together, and a neutral-labels-only
  baseline of 0.45.
• It covers computeMetadataFeasibility for: a non-finite nowMs (returns 0), a title of length
  >= 8, a title of length 4-7, and a title of length < 4.
• It covers opportunityMetadataInternals.titlesOverlap for: an empty left, an empty right, exact
  equality, a containment with shorter.length >= 12, a containment with shorter.length < 12,
  and left-longer vs right-longer (both swap arms).
• It covers computeMetadataDupRisk for: a blank title (returns 1), a self-peer skip, a
  cross-repo peer skip, a same-repo overlapping peer, and a zero-overlap batch (returns 0).
• It covers buildMetadataRankInput with highRiskDuplicateClusters and openPullRequests both
  present and both absent (the two bare-suppressed ?? 0 arms at lines 204 and 206), and with
  updatedAt/createdAt both present and both absent (the bare-suppressed arm at line 215).
• npm run test:coverage reports no uncovered line or branch in
  packages/loopover-engine/src/opportunity-metadata.ts outside the retained internals block.

All Deliverables above are required in a single PR. A PR that satisfies only some of them — for example
removing the directives and adding tests for computeMetadataPotential only, leaving
computeMetadataDupRisk's five branches uncovered — does not resolve this issue.

Test Coverage Requirements

packages/loopover-engine/src/**/*.ts is inside coverage.include in vitest.config.ts and
carries its own engine Codecov flag, so the 99%+ branch-counted codecov/patch gate applies to every
line this PR touches — which, because the touched lines are precisely the ones the directives were
hiding, means the gate itself enforces the tests above. Both arms of every un-suppressed conditional
need a test; the enumerated cases in the Deliverables are the minimum set. Measure with
npm run test:coverage (unsharded) before pushing, per the contributor guide.

Expected Outcome

The metadata ranker's scoring branches are genuinely branch-covered and any future change to a
threshold, weight, or filter is caught by a test, instead of being invisible behind 18 coverage
suppressions in a file the coverage gate is supposed to be grading.

Links & Resources

• packages/loopover-engine/src/opportunity-metadata.ts (whole file)
• packages/loopover-engine/test/opportunity-ranker.test.ts,
  packages/loopover-engine/test/opportunity-competition.test.ts (the suite style to mirror)
• vitest.config.ts coverage.include; codecov.yml flags.engine

[linear-code]

JSO-6690