xl: Introduce new blocking writer to make CreateFile atomic. (#1362)

Creates a new write closer that must be released
by the read consumer. This is necessary so that
while commiting the underlying writers in erasure
coding we need to make sure we reply success only if
we have committed to disk.

This in turn also fixes plethora of bugs related to
subsequent PutObject() races with namespace locking.

This patch also enables most of the tests, other than
ListObjects paging which has some issues still.

Fixes #1358, #1360

object-api-multipart.go

@@ -289,10 +289,6 @@ func (o objectAPI) NewMultipartUpload(bucket, object string) (string, *probe.Err
 			// uploadIDPath doesn't exist, so create empty file to reserve the name
 			var w io.WriteCloser
 			if w, e = o.storage.CreateFile(minioMetaVolume, uploadIDPath); e == nil {
-				// Just write some data for erasure code, rather than zero bytes.
-				if _, e = w.Write([]byte(uploadID)); e != nil {
-					return "", probe.NewError(toObjectErr(e, minioMetaVolume, uploadIDPath))
-				}
 				// Close the writer.
 				if e = w.Close(); e != nil {
 					return "", probe.NewError(e)