Stellar Index v0.27.0

[v0.27.0] — 2026-08-08

Added

• stellar.contract_active_ledgers — the quiet-contract index.
  Cold /v1/contracts/{id} for a low-activity contract scanned the
  whole events key range backwards (avg 9.3s / 114M rows — the
  inverse of the v0.26.1 busy-contract pathology; three user-reported
  503s). New per-(contract, ledger) activity index
  (deploy/clickhouse/contract_active_ledgers.sql, ~20-40 GiB vs
  ~800 GiB for a per-event index): the reader walks the contract's
  recent active ledgers by primary key (µs) and bounds the events
  read to them — lossless for the page and its cursor, and an empty
  walk is an authoritative "no events". Deliberately count-free so
  re-derives/replays can't double-count (the migration-0059 class);
  kept current by a materialized view, with
  stellarindex-ops ch-contract-ledgers-backfill (windowed,
  resumable) covering history. Reader falls back to the unbounded
  walk when the index is absent or empty (requireRows probe — an
  MV-drop must not read as "no contract has events").

Fixed

• Capacity errors never render as 500. /accounts/{g}/activity
  served "detached refresh capacity saturated" as a 500 Internal
  error, and /operations did the same with driver-level ClickHouse
  i/o timeouts. All 14 explorer handler sites now classify through
  one retryableColdMiss (extended to net.Error timeouts) → the
  truthful retryable 503.