Fix memory leaks in server disposal

[RicoSuter]

Two related disposal bugs prevent the server object graph from being garbage collected after StopAsync/Dispose when clients were connected. Discovered during resilience testing of a long-running OPC UA server with repeated restart cycles (~2 MB leaked per cycle).

Disclaimer: This bug has been found with hours of manual work with AI and my local OPC UA stress test application and should be safe - please review the changes as I don't really know the OPC UA code base in depth.

1. UaSCUaBinaryChannel.Dispose() does not dispose the Socket property

Pending async I/O keeps the socket alive in SocketAsyncEngine, which retains the entire channel/listener graph and prevents GC. Fixed by adding Utils.SilentDispose(Socket) in Dispose.

Double-dispose is safe because TcpMessageSocket.Close() nulls m_socket before disposing, making subsequent Dispose() calls a no-op. TcpServerChannel.Dispose() acquires DataLock, preventing races with ForceChannelFault/SendErrorMessage.

2. StandardServer.OnServerStarted() subscribes CertificateValidator.CertificateUpdate but never unsubscribes

Since the CertificateValidator is shared (from ApplicationConfiguration) and outlives the server, the delegate retains every disposed server instance. Fixed by unsubscribing in StandardServer.Dispose(bool) inside the disposing guard.

GC retention chains (before fix)

SocketAsyncEngine → Socket → TcpMessageSocket → TcpServerChannel
  → TcpTransportListener → m_callback → SessionEndpoint → Server

Socket → Channel → ChannelQuotas → CertificateValidator
  → CertificateUpdateEventHandler → Server

Verification

Tested in a long-running OPC UA server with chaos testing (repeated server stop/start cycles with connected clients):

• Before fix: ~2 MB/cycle heap growth, retained server instances accumulating
• After fix: HeapMB stable (75–77 MB range over 22+ cycles), no retained server instances

Notes

• Both fixes are inside existing Dispose(bool) methods, inside the disposing guard
• No behavioral changes during normal operation — only affects cleanup after disposal
• These bugs only manifest when servers are repeatedly created and destroyed within the same process (e.g., server restart loops). In single-lifecycle production deployments, process exit cleans everything up.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 64.68%. Comparing base (d25caff) to head (3e9e183).
⚠️ Report is 121 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #3560       +/-   ##
===========================================
+ Coverage   51.86%   64.68%   +12.81%     
===========================================
  Files         370      458       +88     
  Lines       78618    96776    +18158     
  Branches    13650    16265     +2615     
===========================================
+ Hits        40779    62601    +21822     
+ Misses      33705    29161     -4544     
- Partials     4134     5014      +880     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.