signals(contributor-pr-monitor): don't flag PRs as duplicate-prone when both titles normalize to empty

[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

src/signals/contributor-open-pr-monitor.ts's duplicatePronePullNumbers groups a repo's open PRs by
a normalized version of their title, and flags every PR in any group of 2+ as duplicate_prone:

function duplicatePronePullNumbers(openPullRequests: PullRequestRecord[]): Set<number> {
  const flagged = new Set<number>();
  const byNormalizedTitle = new Map<string, PullRequestRecord[]>();
  for (const pr of openPullRequests) {
    const key = normalizeTitle(pr.title);
    const bucket = byNormalizedTitle.get(key) ?? [];
    bucket.push(pr);
    byNormalizedTitle.set(key, bucket);
  }
  for (const bucket of byNormalizedTitle.values()) {
    if (bucket.length < 2) continue;
    for (const pr of bucket) flagged.add(pr.number);
  }
  // ...
}

function normalizeTitle(title: string): string {
  return title
    .toLowerCase()
    .replace(DRAFT_TITLE_PATTERN, "")
    .replace(/^wip:\s*/i, "")
    .replace(/[^a-z0-9]+/g, " ")
    .trim();
}

normalizeTitle strips every character that isn't [a-z0-9] down to a single space, then trims. A PR
title made up entirely of punctuation, emoji, or other non-alphanumeric characters — e.g. "...",
"!!!", "---", "🎉🎉🎉" — normalizes to the empty string "". Any two (or more) open PRs on the
same repo whose titles both happen to normalize to "" collide into the same map bucket and get
flagged as duplicate_prone, even though their real titles have nothing in common and the PRs may be
about completely unrelated changes.

This flag is not cosmetic-only: duplicatePronePullNumbers's output feeds
mapPendingClassToWorkClassification → nextStepsForClassification (same file), which tells the
contributor to check the flagged PRs for overlap and "close or consolidate duplicates" — a misleading
and potentially harmful suggestion for two genuinely unrelated PRs that only coincidentally share an
empty normalized title.

No existing test in test/unit/contributor-open-pr-monitor.test.ts covers this case — the file's
duplicate-detection tests only exercise real-word title collisions and the draft-marker-stripping
regression.

Requirements

• duplicatePronePullNumbers must never flag two PRs as duplicate-prone solely because both of their
  titles normalize to the empty string. A normalized title of "" must be excluded from the
  duplicate-grouping logic entirely (it must not be treated as a valid grouping key that two or more PRs
  can collide into).
• This must not change behavior for any pair of PRs whose titles normalize to the same NON-EMPTY string
  — that is the intended, working case and must continue to flag as duplicate_prone exactly as today.
• The separate wip/duplicate-label-based flagging in the same function (the wip block below the
  title-grouping loop) must be unchanged.

Deliverables

• duplicatePronePullNumbers (or normalizeTitle's caller) skips the empty-string normalized-title
  bucket when deciding which PRs to flag, verified by a new test asserting that two open PRs titled
  "..." and "🎉🎉🎉" (or any two punctuation/emoji-only titles that both normalize to "") are
  NOT included in the returned Set<number>.
• An existing-behavior regression test confirms two PRs with the SAME non-empty normalized title
  (e.g. both titled "Fix bug" and "fix bug!!", which normalize to the same non-empty string)
  are still correctly flagged as duplicate-prone after this fix — the fix must not weaken the real
  detection case.

Both Deliverables are required in the same PR.

Test Coverage Requirements

This repo's Codecov patch gate requires 99%+ patch coverage on every changed line and branch under
src/**. src/signals/contributor-open-pr-monitor.ts is inside src/**. Both new/updated tests above
must exercise the actual empty-normalized-title collision (not just assert on normalizeTitle's return
value in isolation) so the fix is verified end-to-end through duplicatePronePullNumbers's real return
value.

Expected Outcome

Two open PRs whose titles both happen to normalize to an empty string (all-punctuation, all-emoji, or
otherwise entirely non-alphanumeric titles) are never flagged as duplicate-prone of each other. A
contributor working on such a PR no longer sees a misleading "check overlap / close or consolidate
duplicates" suggestion generated purely from an empty-string coincidence.

Links & Resources

• src/signals/contributor-open-pr-monitor.ts — duplicatePronePullNumbers and normalizeTitle
  (around lines 237-262).
• test/unit/contributor-open-pr-monitor.test.ts — the existing duplicate-title test suite to extend.

[linear-code]

JSO-7329

[aryansk]

I’m working on this with a narrow change to exclude only the empty normalized-title bucket while preserving real non-empty duplicate-title detection and label-based behavior. I’ll add the requested end-to-end regression cases and run the focused contributor-PR-monitor tests before opening the PR.