Global failures meter

[jkozlowski]

What happened?

Both MetricsInvocationEventHandler and TaggedMetricsServiceInvocationEventHandler use the same metric name, but potentially different metric registries (dropwizard MetricRegistry vs TaggedMetricRegistry). Depending on how metrics are reported from those 2, I believe there's a potential that the numbers reported to monitoring infra are non-deterministic?

What did you want to happen?

https://github.com/palantir/atlasdb has those classes somewhat copied, and therefore suffers from the same issue. I am curious what the correct behavior here should be.

[jkozlowski]

@carterkozak @iamdanfox for SA