fix: discard auth migration telemetry silently in broker mode

[clechevallier]

In broker mode (serverUrl == serverUrlV2), _runnerServer.ConnectAsync is intentionally skipped in BrokerMessageListener.CreateSessionAsync since only the broker connection is needed. However, ReportAuthMigrationTelemetryAsync still calls UpdateAgentUpdateStateAsync which calls CheckConnection(Generic), throwing InvalidOperationException: SetConnection Generic.

The exception is caught, but the telemetry is re-enqueued for retry, creating a silent infinite loop that emits an ERR every ~60 seconds for the entire lifetime of every runner pod using broker mode (e.g. ARC with GitHub App auth).

[RUNNER 2026-03-13 15:39:55Z ERR  Runner] Failed to report auth migration telemetry.
[RUNNER 2026-03-13 15:39:55Z ERR  Runner] System.InvalidOperationException: SetConnection Generic
[RUNNER 2026-03-13 15:39:55Z ERR  Runner]    at GitHub.Runner.Common.RunnerServer.CheckConnection(RunnerConnectionType connectionType)
[RUNNER 2026-03-13 15:39:55Z ERR  Runner]    at GitHub.Runner.Common.RunnerServer.UpdateAgentUpdateStateAsync(Int32 agentPoolId, UInt64 agentId, String currentState, String trace, CancellationToken cancellationToken)
[RUNNER 2026-03-13 15:39:55Z ERR  Runner]    at GitHub.Runner.Listener.Runner.ReportAuthMigrationTelemetryAsync(CancellationToken token)

Fix by catching the specific InvalidOperationException from CheckConnection before the generic catch, and discarding the telemetry entry with a Trace.Verbose instead of re-enqueuing. This breaks the infinite retry loop with zero impact on non-broker runners where _runnerServer is always connected.