Extend should erase_on_boot

[jbspeakr]

Currently when extending an existing ring, all user-data gets copied to new instances, including erase_on_boot set to false. this needs to mounting issues during taupage boot, as the new volumes need to get formatted initially.

I would expect this to work slightly different: extending a ring should lead to a status where the new ring is able to operate. hence, this requires initially formatting the nodes and setting erase_on_boot to false after cassandra was setup up correctly.

[smirnov]

(Erase on boot should not be added to the metadata alltogether, then the format will be done by taupage once, based on a tag of a volume and this will not require an extra restart.