Metrics Collector: Fix adoption stats

[and-rewsmith]

We removed the test specific logic from the metrics collector. In doing so, this removed a get twin call that was relied on for gauging usage stats.

The solution is to recreate the ModuleClient (after product info and potentially all other fields are set) even in the Log Analytics code path. This change needs some synchronization as to not interfere with metrics potentially being published from the IotMessage metrics upload path (which also uses ModuleClient).

To achieve this synchronization, I chose to create a wrapper around the ModuleClient which serves as the shared reference across multiple Tasks. Multiple of these Tasks might use the ModuleClient. One Task could be recreating it while the other could be attempting to publish messages. In order to achieve synchronization I chose to use a SemaphoreSlim because I needed some async lock.

I ran into some issues when attempting to recreate the ModuleClient.

1. If messages tried to be published too quickly, the initial message send would hang.
2. When calling CloseAsync() the SDK threw an exception. (SDK bug imo)

I managed to resolve these issues by not calling CloseAsync() and waiting 1 minute to publish messages. I confirmed in EdgeHub logs that multiple connections did not exist.

[damonbarry]

Can you describe (maybe in the PR description) what you're achieving by using a semaphore?