fix: possible SIGILL and memory leak

[Henje]

This PR fixes two bugs which I found while "stress-testing" the library with

var fsevents = require("fsevents");
while(true) {
  new fsevents.FSEvents("/", () => {});
}

and

var fsevents = require("fsevents");
while(true) {
  fsevents("/");
}

The first one is a SIGILL signal generated in pthread_mutex_destroy because the mutex is not always initialized. This is caused by not initializing lockStarted which causes it to sometimes be true, depending on where the object was allocated. A simple fix is to initialize it to false.

The other bug is a cyclic dependency which is hidden from v8 and therefore never freed. If the closure handed to FSEvents has a reference to the event and FSEvents saves the callback internally, a cyclic dependency is created. The problem arises, because v8 can only directly see the closure->FSEvents dependency not the internal one. My fix is to "promote" the closure reference to the associated v8::Object, therefore allowing v8 to see the dependency. Before fixing this bug I had no prior experience with v8 so my solution might not be ideal.

In general I did not see any contribution guidelines so I would be happy to amend these changes where necessary.

[bnoordhuis]

Left some comments but mostly LGTM.

Thanks for doing this and sorry for the delay, I'm afraid the notification got buried.

[bnoordhuis]

Can you make this explicit?

[bnoordhuis]

Can you do this in the constructor? This is a C++11 construct and we (nominally) still support Node.js v0.10 and v0.12, which are C++03.

[bnoordhuis]

There are Nan methods for this (Nan::Get() and Nan::Set().)

[bnoordhuis]

Can you use Nan::To<v8::Function>(...)? It does an IsFunction() check before casting.

[bnoordhuis]

Not necessary to cast to v8::Function.

[Henje]

Thank you for you feedback. The C++11 initialization must have slipped through. I hope my changes satisfy your review.

[bnoordhuis]

Thanks, LGTM.