As the summary says, we should allow jobs to complete when some targets are unavailable.
Stream sessions can only fail via the FailureDetector
sstableloader should detect and report failures
Patch requires CASSANDRA-4146. It would be nice if we could support the notion of consistency level here, but it's not as easy as setting a ConsistencyLevel directive like other output formats, so instead we just allow defining a maximum number of hosts that are allowed to fail before the entire job fails, and default that to zero.