Implement fast uploader

[milliondreams]

Currently snackfs relies on HDFS 'copyFromLocal' or equivalent command to send data to snackfs. This command doesn't understand snackfs internals and treats it as just another (HDFS compatible) filesystem, writing chunks of bytes to output stream.

Since snackfs is a fat client and all the logic to create inodes and blocks/subblocks from the file resides on the client, we can use it improve upload performance by splitting the files into chunks equal to the "subblock" size and writing multiple subblocks to different C* nodes in parallel.

Once all the subblocks are written, we can update the inode in one go.