fix flaky Microsoft.AspNetCore.SignalR.Client.Tests.TestServerTests.WebSocketsWorks

[DeagleGross]

The WebSocketsWorks() and LongPollingWorks() tests had a race condition where implicit using/await using disposal could dispose the host's IServiceProvider while the server-side OnDisconnectedAsync was still running, causing ObjectDisposedException.

Fixes #65914

[Copilot]

Pull request overview

Fixes flakiness in SignalR client TestServerTests by ensuring the in-proc TestServer host isn’t disposed while server-side OnDisconnectedAsync is still running, eliminating ObjectDisposedException errors and unquarantining the affected tests.

Changes:

• Removed quarantine annotations from WebSocketsWorks() and LongPollingWorks().
• Added explicit connection.StopAsync() and host.StopAsync() to drive a graceful shutdown order before disposal.

[Copilot]

connection.StopAsync() / host.StopAsync() are only executed on the success path. If an earlier await/Assert throws, cleanup falls back to disposal again and can reintroduce the same race (and can also cause additional error logs during StartVerifiableLog() disposal). Consider moving the explicit stop calls into a finally so they run regardless of test failure.

[Copilot]

Same as above: these explicit stop calls only run on the success path. To avoid disposal-time races/log noise when the test fails early, wrap the body in a try/finally and perform connection.StopAsync() / host.StopAsync() in the finally (guarding for partially-initialized state as needed).

[BrennanConroy]

I don't think this change does anything. We already have await using var connection = ...; which will close the connection before closing the host.

I'm guessing an actual server-side change is needed to improve the lifecycle handling of a closing server.