Instability in the watch test: Windows

[promiseofcake]

Reported via @fruch in #65

there is some instability in the watch test (in windows it's failing constently)

We are not testing for Windows, an improvement would be to address that.

[fruch]

I have a bunch of fixes for windows I'll put them into a PR,
anyhow the instability I was refering too, was that it failed in the travis CI:
https://travis-ci.org/vsco/dcdr/builds/352241579?utm_source=github_status&utm_medium=notification

[fruch]

I think I've pined down the issue.

There two call to watch, one on New and one on Watch. there a lock on c.watcher.New() and also on c.watcher.Watch().

I'm a total newbie on golang, but seem like a race condition that makes the test pass on linux (and because fsnotify works different on windows, it's get stuck always, and the bug in the test is surfaced)

[fruch]

I think this change is the cause:
d8d6f27

[fruch]

I'm quite sure it's not windows only this issue, seen it fail on CI on #71 (rerun magicly solved it)