fix(contributor-card): retry failed fetches instead of hiding the card forever

[rrbe]

Symptom

The contributor card sometimes never appears for a user (no card, not even a loading skeleton) and stays gone until a full page reload. It looked page-specific — e.g. "works on the PR detail page, but not on the PR list" — but isn't.

Root cause

fetchContributorInfo returns null on any error (rate limit, network blip, service-worker hiccup). sync() then cached that null:

infoCache.set(cacheKey, data ?? null);

and the cache-hit branch renders nothing for a null entry and never refetches:

if (infoCache.has(cacheKey)) {
  existing?.remove();
  const info = infoCache.get(cacheKey);
  if (info) placePanel(card, buildPanel(login, info)); // null → nothing, forever
  return;
}

So a single transient failure permanently suppressed the card for that user for the rest of the session. The "PR list vs detail" pattern was just whichever hover happened to poison the cache first (the cache key is repo#login, page-independent).

Fix

• Cache successful fetches only (infoCache: Map<string, ContributorInfo>).
• Record failures in failedFetchAt with a timestamp and retry after a 60s cooldown (and on any fresh load — the maps are in-memory). The cooldown stops a busy hovercard from re-fetching on every DOM re-render, while still recovering on a later hover.
• cleanupContributorCard() clears the new map too.

Verification

• Live (Chrome, real hover on the PR list page): before the fix, CodyTseng's card stayed absent for 3s+ (no skeleton). After clearing the in-memory cache (page reload), the same hover fetched and rendered fine (skeleton → real card in ~1.6s) — confirming the fetch works and the poisoned null cache was the cause.
• Added a regression test: a failed fetch shows nothing, does not refetch within the cooldown (no network storm), and retries after the cooldown — then renders the card.
• pnpm test → 227 passing. pnpm build clean.