Failures in after_success don’t actually fail the build.

Remove the cli before test

This test is super flakey for some reason, and `before` hooks also
appear to be get retried by mocha.

The was only required for when testing against LibSass < 3.3 which
is no longer the case. As such this hook serves no purpose.

Don't run tests when building a Windows release

Fixes to troubleshooting doc

package.json will not be created without npm init and a --save is required to append the installed version of package onto the package.json

Move the two CLI Usage into separate lines

Publish tests to npm

Add a nice little progress bar while downloading binary

Increase the binary fetch request timeout

I found the 1s timeout to be really fragile when testing locally.
The intent of the timeout is to prevent the install process hanging
indefinitely. It makes sense to be very generous so this increases
the timeout to 60s.

The timeout applies to both the initial connection as well as
individual packets.

Forked from #1649.

This PR adds a small ascii progress bar to the binary download.
Stalled installations make up a steady background noise of our
issues. It's my hope that by making the binary fetch more
transparent we can stem the tide of those issues.

In order to address @saper's [concerns][1] this progress bar will
respect npm's `progress` config flag.

Downloading
```
Start downloading binary at https://github.com/sass/node-sass/releases/download/v3.8.0/darwin-x64-48_binding.node
Total 2602136 [===============          ] 1566256 60% 4.9s
```
Success
```
Start downloading binary at https://github.com/sass/node-sass/releases/download/v3.8.0/darwin-x64-48_binding.node
Total 2602136 [=========================] 2602136 100% 0.0s
Binary downloaded and installed at /tmp/node-sass/vendor/darwin-x64-48/binding.node
```

[1]: https://github.com/sass/node-sass/pull/1649#issuecomment-242720257

Retry CLI specs that fail three times

For some reason our test fail due to timeouts on seemingly random
cli tests. This instability has been growing worse. I've spent the
better part of the last two days looking into the issue without a
resolution. I'm starting to think it's an artefact cause by the
virtualisation used in CI, potentially related to resources
starvation.

The specs that fail do so trying to spawn a child process. The
process never spawns eventually resulting in a timeout. This is why
@nschonni attempt at addressing this by increasing the timeout in

Since the spawn failure is random, and catastrophic the best solution
I can think of in the short-medium term is to instruct mocha to retry
failing tests individually a couple times.

I've scoped this behaviour to the cli specs because at the moment
we have some specs that are not reentrant.

This PR adds a timeout to the binary fetch request as well as a
useful error message.

It came to light in #1690 that the download can block for a long time.

```
npm info lifecycle node-sass@3.8.0~install: node-sass@3.8.0

> node-sass@3.8.0 install /Usersproject-path/node_modules/node-sass
> node scripts/install.js

Build timed out (after 35 minutes). Marking the build as failed.
```

We see a fair amount of hung installer issues that could be related
to this. The error output will give us another data point in those
cases.

This PR adds a timeout to the binary fetch request as well as a
useful error message.

It came to light in #1690 that the download can block for a long time.

```
npm info lifecycle node-sass@3.8.0~install: node-sass@3.8.0

> node-sass@3.8.0 install /Usersproject-path/node_modules/node-sass
> node scripts/install.js

Build timed out (after 35 minutes). Marking the build as failed.
```

We see a fair amount of hung installer issues that could be related
to this. The error output will give us another data point in those
cases.

This is required for nodejs/citgm#165.

I conceed it will increase the size of the tarball. This isn't ideal
given the background noise of installation issues atm, but IMHO the
value of being in the CITGM is worth it.

The additional transparency into the binary download added by #1649
will go a long to alleviating the addition tarball size.

- New function buildIncludePaths to build the includePaths string for passing through to libsass
- In buildIncludePaths, check for SASS_PATH environment variable, and if found include them behind any other specified include-paths
- Add two tests to `.render(options, callback)`:
  - "should check SASS_PATH in the specified order"
  - "should prefer include path over SASS_PATH"

libsass_ext is not defined

Troubleshooting: Check COMSPEC for Windows

Looks like I accidentally reverted my fix while rebasing
d6cd0134

ESLint conversion from JSHint

Fixes: https://github.com/sass/node-sass/pull/1631
Fixes: https://github.com/electron/electron/issues/6462
Broken-By: https://github.com/sass/node-sass/pull/1101
Fix-Provided-By: https://github.com/sass/node-sass/pull/1631

http://eslint.org/docs/rules/no-unused-vars

http://eslint.org/docs/rules/no-unused-expressions

Suppressed the ones coming from the Sass Types

Newer rule system and auto-fixing errors

We don't use gitter anymore,