semver: do not support upload() from url

[fhinkel]

Remove the upload() from url functionality. Instead, use
File.createWriteStream.

For performance reasons, nodejs-storage should not require the request
module. As we are currently exposing request as part of the public
API for upload(), not using request is a breaking change. Instead of
using a different dependency for the http requests, we leave this decision
to the users.

If you were using upload() from a url, replace it with the following:

const request = require('request');
let readStream = request(url).createReadStream();
const file = bucket.file(name)
const writeStream = file.createWriteStream();
readStream.pipe(writeStream)

Use node-fetch instead of request in system tests.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@e68242a). Click here to learn what that means.
The diff coverage is 100%.

@@           Coverage Diff           @@
##             master   #337   +/-   ##
=======================================
  Coverage          ?   100%           
=======================================
  Files             ?      7           
  Lines             ?   1016           
  Branches          ?      0           
=======================================
  Hits              ?   1016           
  Misses            ?      0           
  Partials          ?      0

Impacted Files	Coverage Δ
src/bucket.js	100% <100%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e68242a...cfaa816. Read the comment docs.

[stephenplusplus]

Why not support it?

[fhinkel]

Sorry, didn't include the reasoning, see updated commit message.

[crwilcox]

LGTM. Before merging I would like to take some time to make sure we are prepared to incr semver-major.

[stephenplusplus]

Heads up @Kiricon-- we're going to be removing the feature you added for us. Do you think it's a mistake? The dilemma is that it forces us to keep a dependency on request, which we are trying to avoid (it's big, and it has introduced some security concerns).

We have a couple of options:

1. Get rid of the feature. The user can handle fetching the data themselves, and pipe it into a file.createWriteStream(). This is definitely not as easy as "put this remote file in my bucket". The user will have to create a File object, and familiarize themselves with the streaming APIs that both our library and their possibly-new HTTP-fetching library.
2. Use a different request library, but somehow remain library-agnostic when accepting common HTTP options, like headers.

Thoughts and ideas welcome!

[fhinkel]

It's been a week with no objections, can we merge this?

[nnmrts]

I consider this as a mistake. There are no "perfomance reasons" to do this. It's now harder to upload files, without any benefits performance-wise. Welp.