fix(deps): follow-up fixes from PR #135 review + quality-gate cycle - #136

Fixes remaining findings from /pr-review + this session's own /quality-gate
on PR gordon-code#135:
- Extracts fetchNodeBodiesBatched<TNode,K,V>() as a shared helper for
  fetchDashboardIssueBodies/fetchDepPRBodies, removing ~20 lines of
  duplicated batching/timeout/error-reporting boilerplate. Both now
  return { bodies, failedIds } instead of a bare Map.
- Adds a 5-minute failure-cooldown Map in DashboardPage.tsx so a
  persistently-failing dependency PR isn't re-fetched (burning a full
  20s timeout + an unbounded Sentry event) every poll cycle — mirrors
  the existing _repoLastTargeted/TARGETED_COOLDOWN_MS pattern in
  poll.ts. The cooldown-prune step runs on every effect evaluation
  (not only when something needs fetching), so a PR that fails,
  leaves the dependency set, and reopens isn't stuck on a stale entry.
- Restores always-on console.debug logging in both files (reverting an
  earlier DEV-gating pass) to match this codebase's dominant
  console.debug/info convention (most existing calls — auth.ts,
  notifications.ts, OAuth callbacks — already ship unconditionally to
  production) and to preserve the original implementation plan's
  production-diagnostic strategy for a still-open reactivity-bug
  investigation.
- Rewords a debug log message that leaked internal authoring
  commentary, and rewrites a test comment that referenced an external
  planning-document section number.
- Adds a banner comment matching api.ts's established section-header
  convention.
- Adds tests: full 7-category sort-order regression, a Digest filter
  functional test, an integration test proving the hung-request guard
  releases and a later poll retries, and three cooldown-specific
  integration tests (same-PR exclusion within the cooldown window,
  cooldown-entry pruning when a PR leaves the dependency set, and
  pruning when no other PR triggers a fetch in between).