storage: use streams.

[stephenplusplus]

I know this is beating a dead horse, but I finally got streams working with the help of duplexify. Getting this in before M1 release would be fantastic, since this is such a fundamental Node philosophy, and is relied on in many apps.

Given that the streams worked out, I simplified the signature of Bucket#write. If a user wishes to use our writable, they are free to open their readable pipeline however they wish, then simply pipe it to bucket.createWriteStream() (as opposed to passing a readable to #write). As an added shortcut, Bucket#write accepts a string or buffer directly: bucket.write('file', 'Hi!', function(err) {});

[rakyll]

I liked this approach much! One question about Bucket.prototype.write: Do we need it at all? createWriteStream(...).write(buffer) isn't working?

[stephenplusplus]

write is just left around for when one needs to write a string or buffer. As far as I know, without our write method, the user would have to design/implement their own string/buffer readable stream in order to be able to pipe it to createWriteStream().

For those use cases, the api now looks like:

bucket.write('destination-file-name', 'Contents', function(err) {});
// or
bucket.write('destination-file-name', {
  data: 'Contents',
  metadata: {
    // ...
  }
}, function(err) {});

It also still handles metadata-only posts when there is no data provided.

[rakyll]

Writable streams provide a write method: http://nodejs.org/api/stream.html#stream_writable_write_chunk_encoding_callback

var st = bucket.createWriteableStream(...);
st.write(buffer);
st.end();

should work. But for metadata only requests, it's weird to require end before writing any data to the stream.

[rakyll]

I'm merging this change, will run some perf test later this week. Much thanks!