Improve snapshot backup/restore test coverage

[gyuho]

Snapshot fetches the point-in-time state of etcd backend database. API itself is simple; stream key-value pairs by iterating each bucket. However, there have been many unexpected behaviors with frequent snapshots and restore.

Increase test coverage around snapshot API and its restore operations. Some of the missing test cases are:

• restore multi-node cluster from same snapshot file
  • *: add snapshot package #9118 snapshot: add "snapshot" restore/save/member-add tests #9198
• do frequent snapshots and make sure db file size remains static
• make embedded etcd return error on initial integrity check rather than os.Exit
  • *: check data corruption on boot #8554 had to test this in e2e because of exit code

More...?