ServiceRadar v1.4.89

This release lets large integration syncs complete. Ingestion batches are
bounded and their deadline scales with batch size, and a batch that loses
an active-IP race inside the identity fence is retried instead of failing
the run.

Upgrade Notes

• The sync ingestion worker deadline now applies per chunk: a batch gets
  sync_ingestor_worker_timeout_ms (default 120 s) for each chunk it
  carries. The setting is configurable through Helm
  core.syncIngestor.workerTimeoutMs or the core environment variable
  SYNC_INGESTOR_WORKER_TIMEOUT_MS; values that are not positive
  integers fall back to the default (PR #5575).

Fixes

• Ingestion batches take at most queue_max_chunks chunks (default 10).
  Before, every batch took all pending chunks and a timed-out batch was
  retried with everything that had arrived since, so large sync runs grew
  past their deadline, were killed on every retry, and left later runs
  marked incomplete. A retry now carries only the chunks that failed
  (PR #5575).
• Device writes inside the identity fence no longer open nested
  transactions on the inventory-rollup bypass and address-release paths.
  An active-IP conflict now reaches the sync ingestor as a retryable
  Postgres error rather than a connection error, so the batch is retried
  instead of rejecting every chunk and leaving the run incomplete
  (PR #5574).

Dependencies

• github.com/spiffe/go-spiffe/v2 2.8.2 fixes a data race reading Workload
  API bundles (PR #4887). github.com/bluenviron/gortsplib/v5 is 5.6.6
  (PR #4890) and golang.org/x/net is 0.59.0 (PR #4416).

Docs and CI

• A CHANGELOG entry added outside a release cut was removed, restoring the
  v1.4.88 notes as released (PR #5576, PR #5577).