[IO-546] ClosedOutputStream#flush should throw

[tjcelaya]

While debugging an issue involving usage of CloseShieldOutputStream I discovered that ClosedOutputStream#flush is not overridden and will silently succeed. Not sure how much of a breaking change this might be but I think it makes more sense for ClosedOutputStream#flush to throw. This is only really meaningful in contexts where multiple streams are being chained together and some of the streams before CloseShieldOutputStream perform buffering, but it would make behavior more consistent for these more complex use-cases.

See IO-546 where @dekobon has attached an example program.

[coveralls]

Coverage increased (+0.06%) to 90.073% when pulling 854ed1d on tjcelaya:fix/IO-546-ClosedOutputStream-flush into 92a07f9 on apache:master.

[PascalSchumacher]

Thanks!