gh-837 "replica waits infinitely if the master has no xlogs"

This patch doesn't fix the bug itself, but simplifies matters
so that there is no more .inprogress xlog files.

.inprogress xlog files were created to start automatically
from a state when the server crashed with an empty
xlog, and couldn't create a new xlog since a file with the
same LSN in the name already existed. To solve this problem,
simply bump LSN at start if the last file in the recovery
list had zero records in it.

Emtpy xlog files still lead to panic if panic_on_wal_error=true
during recovery.