Measure and improve startup and shutdown times

This is a part of ZDT (zero downtime) version upgrades.

The current version upgrade process is not zero downtime. All nodes are shut down for a while and restarted in a new product version. This takes some time, however:

* It isn't clear, what is actual down time. It includes both shutdown time (after the old process stops accepting/processing transactions) and startup time. Is it minutes? Seconds? Sub-seconds?
* We don't know where exactly time is spent at shutdown and at startup

This ticket is to have more details (add logs, metrics, etc.) about CN shutdown and startup at all levels. It includes, but not limited to, execution layer (new user transactions are no longer accepted), consensus (what happens after the freeze round), Java (JVM startup time, class loading and initialization), database (MerkleDb loading indices, state is restored from a snapshot), PCES replay, and more.