Give a chance for the programs use webtorrent cli to run scripts when download done or exit.

This reverts commit af0929b.

Part of an effort to share more code with `torrent-stream` by
@mafintosh:

mafintosh/torrent-stream#133
mafintosh/torrent-stream#132

Storage is now based on:

https://www.npmjs.com/package/fs-chunk-store (in node)
https://www.npmjs.com/package/memory-chunk-store (in the browser)

and:

https://www.npmjs.com/package/immediate-chunk-store
https://www.npmjs.com/package/torrent-piece

This is a BREAKING change, since users who pass in custom storage
engine with `opts.storage` will need to update their storage backend to
be `abstract-chunk-store` compliant:

https://www.npmjs.com/package/abstract-chunk-store

There are also likely other miscellaneous BREAKING API changes, since
this was a huge refactor of the codebase.

Not really sure what this code was doing — maybe trying to allow a
Torrent object to be passed in to another client? Since
torrent.parsedTorrent is gone and the data just lives directly on the
torrent object, removing this code should work just fine.

The previous behavior of `torrent.downloaded` is now
`torrent.received`. This is bytes received from peers (including
invalid data).

The new `torrent.downloaded` is what most people expected it to be all
along. Bytes completed (excluding invalid data).

This is a lot harder to visualize with the new torrent-piece backend,
since we don’t keep an explicit array of which blocks are reserved.
Just remove it for now.

If we need this visualization back to debug something, it should be
possible to recreate it from torrent-piece with a bit of work.

Add test for private torrent - should disable DHT

Fail public torrent test when using a private torrent

standard

Disable DHT for the torrent's swarm and discovery.

Looks like it was forgotten for 'img' and 'iframe'.

See #429 (comment)

See #429 (comment)

> This is because a windows system doesn't use the shebang so it might
> not know how to execute a js file. Spawning using node and node will
> understand how to read this.

http://blog.travis-ci.com/2013-12-05-speed-up-your-builds-cache-your-dependencies/

This allows Windows to run the tests, as promised in
#429 (comment)

Note that there are a couple of outstanding failures on Windows:

    # client.seed: filesystem path to folder with one file, string
    not ok 8 should be equal
      ---
        operator: equal
        expected: '3a686c32404af0a66913dd5f8d2b40673f8d4490'
        actual:   'e6887b78d89d995876c7cef3476b6fb32c4c4e3d'
      ...
    not ok 9 should be equal
      ---
        operator: equal
        expected: |-
          'magnet:?xt=urn:btih:3a686c32404af0a66913dd5f8d2b40673f8d4490&dn=folder&tr=udp%3A%2F%2Ftracker.webtorrent.io%3A80'
        actual: |-
          'magnet:?xt=urn:btih:e6887b78d89d995876c7cef3476b6fb32c4c4e3d&dn=folder&tr=udp%3A%2F%2Ftracker.webtorrent.io%3A80'

The specs say that the second Blob constructor argument is optional (http://www.w3.org/TR/FileAPI/#constructorParams), however Chrome and FF implement that differently. FF checks if the second arg is not undefined, while Chrome checks "arguments.length".

As a result, if the second argument is undefined, Chrome will throw an error (https://chromium.googlesource.com/chromium/blink/+/72fef91ac1ef679207f51def8133b336a6f6588f/LayoutTests/fast/files/blob-constructor.html#55), while FF will just ignore it.

This change makes it so that the second arg is not passed at all when it's undefined, and Chrome is happy as a result.

This forces Windows checkouts to keep LF line endings for test files.

https://help.github.com/articles/dealing-with-line-endings/#example

resolves #429

…ithub.com/solderzzc/webtorrent into add-test-script-for-multiple-file-webseed