bigquery: managedwriter: increase operational visibility when using in-client retries

**Is your feature request related to a problem? Please describe.**
We frequently see bursts of append rows request retries and errors. When using the built in retry mechanism, we don't receive any indication of what the errors actually are, they are hidden by the client.

**Describe the solution you'd like**
A high level metric giving the reason for the failure (status.Codes level maybe) to help diagnose the cause of the errors.