Update pathwatcher to 6.x

[winstliu]

Should allow text-buffer to compile on Node 4.

Closes #98

[mnquintana]

I'm wondering if we should add Node 4.x to the build matrix maybe (for our Node libraries)? (so we don't necessarily have to rely on user reports)

/cc @zcbenz @atom/feedback

[winstliu]

script/cibuild still uses Node 0.12 though - should I also update that to Node 4?
EDIT: Ninja'd 😆

[zcbenz]

I'm wondering if we should add Node 4.x to the build matrix maybe (for our Node libraries)? (so we don't necessarily have to rely on user reports)

Sounds good to me.

[winstliu]

Not sure where to proceed from here...Clang actually seems to be segfaulting when running under Node 4. There's also a lot of failures related to temp files.

[YurySolovyov]

@50Wliu segfaults may gone after npm rebuild or just rm -rf node_modules + npm install

[winstliu]

Once I get atom/node-pathwatcher#90 all set up and merged, then I'll rebase/update this PR accordingly.

[winstliu]

Tests continue to fail but the install no longer segfaults. npm install also fails on the first run, just like atom/node-pathwatcher (but this time with more errors).
Looking into the spec failures now...

[winstliu]

Ok so I've fixed most of the errors by simply updating temp but there are still three failures related to temp file modification that I don't know how to solve:

TextBuffer
  serialization
    when the buffer has a path
      when the serialized buffer had no unsaved changes
        it loads the current contents of the file at the serialized path
          Expected 'words!' to be 'words'. (spec/text-buffer-spec.coffee:913:39)
  ::onDidChangePath()
    it notifies observers when the buffer's file is moved
      timeout: timed out after 5000 msec waiting for buffer path change
  when the buffer's file is deleted (via another process)
    when the file is not modified
      it retains its path and reports the buffer as not modified
        Expected true to be falsy. (spec/text-buffer-spec.coffee:1169:45)

[zcbenz]

It seems that buffer.testSerialization doesn't keep the undo stack but the test assumes so, I'm not sure of the expected behavior, /cc @maxbrunsfeld for help.

[maxbrunsfeld]

@zcbenz I think this is due to a pathwatcher bug that I had fixed on the four and five branches, but I neglected to fix on master. I have merged my fix into master and upgraded pathwatcher on this branch.

[zcbenz]

@maxbrunsfeld Thanks!

For the rest of the failing specs, it is actually caused by switching to Linux in Travis CI, it seems that they never passed on Linux in previous versions.

I'm going to suppress those specs for now so we can focus on updating to latest Electron in Atom.