Fix heap corruption issue on windows

[kunalspathak]

In EIO_AfterWatchPort, serialport creates a v8::ArrayBuffer using Nan::NewBuffer() API which creates v8::ArrayBuffer with ArrayBufferCreationMode::kInternalized mode. This means that v8 engine takes the responsibility of freeing the external buffer passed to the v8::ArrayBuffer.
However EIO_AfterWatchPort deletes and frees the buffer at the end of method.

cleanup:
    delete data;
    delete req;

When ArrayBuffer object is garbage collected, it tries to free the external buffer passed to it, but since it is already freed, we get heap corruption and the stack looks pretty much like one shown in #617.
The repro for this is inconsistent as it depends on state of memory at the time of GC, but with Microsoft's AppVerifier tool, there is a consistent repro for this.

The fix is to pass FinalizerCallback to Nan::NewBuffer() API that will internally create the ArrayBuffer in ArrayBufferCreationMode::kExternalized mode. This means that engine won't take the responsibility of freeing the buffer and when collection of arraybuffer happens, it will call the callback which will free the buffer.

Fixes #617, #618

[ghost]

Just checked out this changes. They actually fix the problem with crashing under Windows.

[techninja]

Yep, just tested, this fixes a pretty nasty anonymous crash in Electron as well!