Improve observability with metrics, Phase 1

[gyuho]

Assume no one's going to look at server logs until it breaks.
Log as few as possible. Instead, use metrics and alerts.

Make sure to get feedback from communities and document any breaking changes.

• Backport critical metrics to v3.1, v3.2, v3.3
• Promote "etcd_debugging" metrics to "etcd"; done via *: promote debugging metrics #9764
• Document metrics lists; done via docs,tools/etcd-dump-metrics: automate metrics documentation #9984
• Cover all warn/debug-level logs with metrics
• Remove/Replace merge logs with metrics (all merge logs are for v2, so need no action)
• Add more metrics (e.g. defrag); done via etcdserver/api/v3rpc: add "etcd_network_server_stream_failures_total" #9760 and etcdserver,mvcc: add more storage layer metrics #9761
• Use structured logging

Next phase to add tracing support.