Don't null out WorkspaceRegistration in Dispose()

[jasonmalinowski]

We had a mutable field that was marked as not-nullable, but was nulled out as a part of Dispose(). If a buffer is disconnected from a workspace on a background thread, we raise the event on the background thread which could race with the unsubscribe: we'd null it out, the event fires and then we null ref.

My fix is just to not null this out. There's zero value to nulling this out, and as we see trying to violate safety protections can only cause problems.

[sharwell]

we raise the event on the background thread

This is not allowed to occur after the call to Dispose(). If there is pending work related to this object, the code that calls Dispose() must wait for this work to complete before calling Dispose().

One of the primary intents of ReferenceCountedDisposable<T> is guarding disposable state that gets used in a concurrent context. The code that makes the callback can take a lease on the disposable state that ensures it will not get released prior to the completion of that callback.

[jasonmalinowski]

This is not allowed to occur after the call to Dispose(). If there is pending work related to this object, the code that calls Dispose() must wait for this work to complete before calling Dispose().

In this case, this class is subscribing to an event from an unrelated component. There's no methods being called on the Disposed object.

[jasonmalinowski]

This PR was obsoleted by #49051.