Add additional self-telemetry about ES retries

[blakebarnett]

For more visibility into how many retries it's taking to flush cernan's log lines buffer. I suppose a counter for any failures would be helpful here too if there's not already one.

[blt]

I believe this can be inferred -- as of this writing -- by summing over all the elastic API errors. We could also add a bit of telemetry for just any failure. @blakebarnett got a preference?

[blakebarnett]

A general error count would be good, but having a breakdown by type/source/sink etc. would be even better :)

[blt]

@blakebarnett fully agreed. Main trick now is cernan's internal telemetry is mostly atomic counters. It's hard to get meta-data associated. That said, the sinks/sources do have counters for these things but they're just bare.