fix(google): bound per-site fan-out so large portfolios don't time out

[KuznetsovRA]

Every per-site fetch was issued with a single Promise.allSettled(pairs.map(...)), i.e. one concurrent HTTP request per property. On an account with ~200 verified sites that saturates the local socket pool — under Docker Desktop the entire batch fails with UND_ERR_CONNECT_TIMEOUT. Because allSettled swallows rejections, the Sites table silently renders zeros and dashes instead of surfacing an error, so the app looks broken with no clue why. fetchDailyBreakdown doubles it: two requests per site, ~400 sockets at once.

Add allSettledLimit() — a worker-pool variant of Promise.allSettled that keeps results positionally aligned and preserves the {status, value|reason} shape, so call sites keep their existing handling. Apply it to the seven per-site/per-URL fan-outs (listSitesWithSummary, fetchPerSiteQueries, fetchPerSitePages, fetchDailyBreakdown, fetchQueryHistory, fetchPortfolioDecay, bulkInspect). Per-account fan-outs are left alone — those are a handful of calls at most.

Default limit is 8. Google's per-user Search Console quota is far higher; the cap is there to protect the socket pool, not the quota.

Measured on a 199-site account (Docker Desktop, macOS):
before — 0 impressions / 0 clicks, every row dashed
after — 7d: 16,109 impressions / 1,788 clicks
28d: 72,932 impressions / 7,078 clicks, page renders in ~8s
Independently confirmed against the API: the same 199 requests run in batches of 10 return 186×200 and 13×403 (unverified sites), 865 data rows — the data was always there, only the fan-out was dropping it.