Live OBS tests in MSTest, coverage gate, and the fixes they found

[Agash]

Moves the live OBS validation out of the example app and into the test project, and raises coverage to 93.9% line / 80.9% branch on net9, net10 and net11. Those numbers combine the unit tests and the live run.

Tests

• The request sweeps now run as MSTest integration tests over JSON and MessagePack. The run-transport-tests command is gone from the example app, and the old hand-configured integration class is removed.
• An in-memory OBS (Fakes/FakeObsServer) speaks the wire protocol, so the handshake, receive loop, reconnects, events, batches and group helpers are tested through the real code without a live OBS.
• Every protocol event is checked to raise its public event.
• Every test name follows {Method}_{Scenario}_{ExpectedResult}.

Library fixes the new tests found

• A rejected identify (close code 4009) is now reported as AuthenticationFailureException and is not retried. Before, a wrong password reconnected indefinitely.
• With auto-reconnect off, a lost connection now stays lost. Before, it reconnected once after every drop.
• CanvasCreated, CanvasNameChanged and CanvasRemoved now fire. The event dispatch table is generated from the protocol, so it can't drift again.
• The record, stream and virtual-camera "and wait" helpers now return null on timeout, as documented. Before, they threw.
• SourceExistsAsync no longer reports a dropped connection or a timeout as the source not existing.
• Logging in the groups and hosting now uses [LoggerMessage].

CI

• The live job runs one framework at a time, since every test host drives the same OBS.
• It gates line (90%) and branch (80%) coverage on every framework. The previous summary step could not run: a string split across lines made the workflow invalid YAML.
• It passes Obs.Required=true, so a missing endpoint or fixture fails the run instead of skipping every live test.