Create a unified error boundary strategy

[KevinMGranger]

We sometimes see exceptions that bubble up to the top of the exporter and crash it. Sometimes it's from code processing a metric, sometimes it's from the client we're using to talk to a certain backend, etc.

We should discuss and find a way to contain these errors. An error with a single metric should be logged, and the metric should be skipped. At the same time, we can't let all unknown exceptions be swallowed, because then we hide ones we need to now about. (No Pokémon exception handling here!)

[mpryc]

Moving to discussion.