Immediately flush fatal exceptions to App Insights

Since App Insights batches-and-flushes telemetry and also has a fire-and-forget mechanism for flushing, we can't be guaranteed that messages have made it to the service before we kill the process.

We may be able to implement something similar to this comment for things like the "Host thresholds exceeded" message: https://github.com/Microsoft/ApplicationInsights-dotnet/issues/407#issuecomment-318208844.

No source file, test, or entry point is named. Start with the linked Application Insights comment and determine the host's fatal-exception telemetry path; done would require a decided approach that guarantees the fatal message reaches the service before process termination.

Written by the indexing model from the issue text.