s3a: set fs.s3a.downgrade.syncable.exceptions = true by default

HADOOP-17597 set policy of reacting to hsync() on an s3 output stream to be one of :Fail, warn, with default == fail.

I propose downgrading this to warn. We've done it internally, after having it on fail long enough to identify which processes were doing either of

• having unrealistic expectations about the output stream (fix: move off s3)
• were using hflush() as a variant of flush(), with the failure being an over-reaction