ServiceRadar v1.4.93

This release adds cancellation for running network scans, prevents queued
sweep results from expiring under burst load, and fixes release publication
when the optional macOS build does not succeed.

Features

• Operators with the networks.sweeps.run permission can cancel a scan
  from Settings > Networks > Running Scans. Confirmation closes the
  execution as failed, then asks its owning agent to stop the group's
  in-flight scan. Offline agents and agents without the sweep_cancel
  capability leave the execution closed and display an explanation.
  Scheduled sweeps continue at their normal interval (PRs #5611, #5612).

Upgrade Notes

• Sweep results now use a bounded background queue with four workers,
  capacity for 128 chunks, a per-agent limit of 32 chunks, and a 32 MiB
  byte budget. Configure item limits and the worker deadline through
  SERVICERADAR_SWEEP_INGESTION_MAX_ITEMS,
  SERVICERADAR_SWEEP_INGESTION_MAX_ITEMS_PER_AGENT, and
  SERVICERADAR_SWEEP_INGESTION_WORKER_TIMEOUT_MS. The existing
  SERVICERADAR_INGESTION_LANE_MAX_BYTES setting controls its byte budget.
  Accepted chunks remain in memory and can be lost if the coordinator
  restarts (PR #5610).

Fixes

• Accepted sweep chunks no longer expire while waiting for ingestion.
  Chunks from one agent and sweep group run in arrival order, while
  other groups share the remaining workers. Full queues reject work
  explicitly so the gateway can retry. Crashed or timed-out workers
  receive one retry; processor errors are final and logged (PR #5610).
• Late sweep chunks still record host results but cannot advance or
  reopen a cancelled, reaped, superseded, or completed execution
  (PR #5612).
• Release asset verification now requires the macOS installer and its
  provenance only after a successful macOS build. Failed, cancelled,
  or skipped optional macOS jobs no longer block publication; all
  other required asset checks remain enforced (PR #5609).