Replace the `split_api_json` Rake task with equivalent logic in 
`gulpfile.js`, so that `npm run gulp` continues to provide the 
end-to-end build process.

Prior to this change, the the 'nanoc:compile' gulp task attempted to 
store the complete output of `nanoc:compile` in a string, which resulted 
in a "stdout maxBuffer exceeded" error. By using `spawn` instead of 
`exec`, we can pipe the output of `bundle exec nanoc compile` to STDIO 
and thus avoid trying to fit it all in a string.

Example build failure: 
https://travis-ci.org/atom/flight-manual.atom.io/builds/550937870#L3656

Fixes the error seen at 
https://travis-ci.org/atom/flight-manual.atom.io/builds/550959348#L7116

Prior to this change, we were getting the following failure on CI:

> No output has been received in the last 10m0s, this potentially
> indicates a stalled build or something wrong with the build itself.
> ...
> The build has been terminated

This was due to the `run_proofer` task running for more than 10 minutes
with no output. To prevent that situation, this commit introduces two
changes:

1. Split up the proofer work. Invoke HTMLProofer on each directory
separately, so that the `run_proofer` task reports intermediate
progress. Instead of waiting until _all_ directories have been proofed
before outputting any results, it will now output results for each
directory immediately after proofing that directory.

2. Don't run HTMLProofer on the API docs. Checking all external links
for all versions of the API docs would likely take a l-o-n-g time. I
don't think we want our builds taking that long. If we _do_ want to run
the proofer on all versions of the API docs, perhaps it could happen as
part of a nightly build, as opposed to happening on every push.

Example build failure: https://travis-ci.org/atom/flight-manual.atom.io/builds/550971951#L7122

Enhance build to process API JSON files for Atom 1.0.0 and newer

Co-authored-by: Antonio Scandurra <[email protected]>

Co-Authored-By: Antonio Scandurra <[email protected]>

Co-Authored-By: Antonio Scandurra <[email protected]>

Co-Authored-By: Antonio Scandurra <[email protected]>

Note: The code samples emitted by this change are not yet rendered with 
syntax highlighted.

This reverts commit 22558dd and adopts an alternative approach to the
problem. We continue to *NOT* run the link checker on the API docs (for
the reasons described in the commit messages for 22558dd), but we run
the HTMLProofer on the root of the output directory, so that internal
links can be properly resolved.

With these changes, we populate the search index with all instance
methods present in the latest version of Atom. Next up, we'll need to
add classes, class methods, and instance properties to the search index.

Co-Authored-By: Antonio Scandurra <[email protected]>

Co-Authored-By: Antonio Scandurra <[email protected]>

- Redirect `/api` to `/api/latest`
- Redirect `/api/latest` to the latest version of the API docs
- Redirect `/api/<some-version>` to `/api/<some-version>/AtomEnvironment`

This is fully functional, but there are a handful of TODO items that I
want to address in some follow-up commits to tidy things up. 😇

Co-Authored-By: Jason Rudolph <[email protected]>

Prior to this change, "Public" classes were labeled as "Public", which
differs from the behavior on atom.io. See screenshot:

https://user-images.githubusercontent.com/2988/60625837-f0bc6d00-9db6-11e9-83f4-0e78c12d96cb.png

This commit resolves that discrepancy. "Public" classes will now be
labeled as "Essential".

Prior to this change, the rendered docs didn't indicate which arguments
were optional. This differed from the behavior on atom.io. See

https://user-images.githubusercontent.com/2988/60625840-f1550380-9db6-11e9-821c-d75d31511e15.png

This commit resolves that discrepancy. Optional arguments will now have
an "optional" badge.

Co-Authored-By: Jason Rudolph <[email protected]>

Co-Authored-By: Jason Rudolph <[email protected]>

With the introduction of kramdown in 73b0e94, we started interpreting
newline characters as hard wraps, which was causing paragraphs to not
take advantage of all the available screen real estate. Documentation is
usually hard-wrapped for readability purposes when navigating source
code, but those concerns don't apply when viewing the same documentation
in a web browser.

This commit fixes the visual glitch illustrated above by supplying
`hard_wrap: false` to  kramdown-gfm. I verified that this doesn't affect
the rendering of other markdown constructs such as code block syntax
highlighting.

This ensures that Safari doesn't apply a custom style that prevents the 
text box from being displayed correctly.