Normalize labels on common parser metrics

[stephen-soltesz]

One day, a single parser service will handle tasks for multiple data types. When that occurs, we want to be able to create a dashboard that distinguishes the behavior for each data type.

This began by adding new "table" labels to several common metrics: https://github.com/m-lab/etl/pull/415

Then generalizing the usage of FileSizeHistogram and RowSizeHistogram to apply to all parsers. https://github.com/m-lab/etl/pull/443 https://github.com/m-lab/etl/pull/448 https://github.com/m-lab/etl/pull/450

There are a few metrics remaining that should be distinguishable.

This work also supports efforts to unify the parser dashboards for monitoring.

[stephen-soltesz]

• etl_task_count - should be labeled with the target table sooner. "unknown" is difficult to work with in the dashboard.
• etl_worker_duration_seconds - should use separate worker targets for separate file types -- this moves the decision to the queue pusher or much earlier in the request handling.
• etl_panic_count - should indicate which parser is generating panics.
• AnnotationRequestCount (and related annotation metrics) - these metrics should be discernible between parsers.