ServiceRadar v1.4.94

This release fixes reply loss and misleading packet-drop statistics in the
agent's raw TCP SYN scanner.

Upgrade Notes

• Raw SYN scans now run one at a time per agent process. Sweep groups
  that start TCP scans together wait for the active scan to finish;
  that wait counts against the scan's context deadline. Streaming
  batches acquire the gate separately, allowing groups to interleave
  between batches (PR #5615).
• Default capture rings now use 64 blocks of 128 KiB instead of eight
  1 MiB blocks. The total remains 8 MiB per ring, with more time for
  readers to consume replies before blocks retire. Explicit ring
  settings still take precedence (PR #5615).

Fixes

• Idle SYN scanners no longer capture another scanner's replies.
  Starting a scan enables reply capture and clears stale ring blocks
  and kernel drop counters before sending probes; teardown restores
  the idle filter. Failure to enable reply capture returns an error
  instead of reporting targets as unreachable (PR #5615).
• Ring sizing no longer loops indefinitely when one block fills the
  available per-ring budget (PR #5615).
• Stored rx_drop_rate_percent now divides dropped packets by received
  plus dropped packets, matching scanner telemetry and correctly
  reporting a scan where every reply was dropped (PR #5615).