Add falback mechanism into instrumentation classes framework for Service Ping

Problem

Service Ping instrumentation classes framework being open to extension by anybody who wants to track some Product Intelligence data is prone to bugs and implementation errors. However it should be resilient as a process and not fail due to single metric being broken. To achieve that fallback mechanism is required, that would rescue any error coming form single metric layer

Recently Service Ping for GitLab.com throws undefined Gitlab::Usage::Metrics::Instrumentations::CountSlackAppInstallationsGbpMetric

Details

The class is found after reconnecting to production rails console.

This could be a case when we deploy while we generate Service Ping.

Investigate if this is a critical issue and if we can fix it

Result

Service ping doesn't fail while deploying