Add a failing test for unhandled rejections and chaining.

Introduces `Q.resetUnhandledRejections`, which is necessary for testing. Setting `Q.unhandledReasons.length = 0` like I was doing is dangerous, since it makes the `unhandledReasons` and `unhandledRejections` get out of sync.

Re: kriskowal#238.

q.js

@@ -750,6 +750,12 @@ function displayUnhandledReasons() {
     unhandledReasonsDisplayed = true;
 }
 
+Q.resetUnhandledRejections = function () {
+    unhandledReasons.length = 0;
+    unhandledRejections.length = 0;
+    unhandledReasonsDisplayed = false;
+};
+
 // Show unhandled rejection reasons if Node exits without handling an
 // outstanding rejection.  (Note that Browserify presently produces a process
 // global without the `EventEmitter` `on` method.)

spec/q-spec.js

@@ -2410,13 +2410,23 @@ describe("possible regressions", function () {
 });
 
 describe("unhandled rejection reporting", function () {
-    it("doesn't report a resolve, then reject (gh-252)", function () {
-        Q.unhandledReasons.length = 0;
+    beforeEach(function () {
+        Q.resetUnhandledRejections();
+    });
 
+    it("doesn't report a resolve, then reject (gh-252)", function () {
         var deferred = Q.defer();
         deferred.resolve();
         deferred.reject();
 
         expect(Q.unhandledReasons.length).toEqual(0);
     });
+
+    it("doesn't report when you chain off a rejection", function () {
+        Q.reject("this will be handled").get("property").fail(function () {
+            // now it should be handled.
+        });
+
+        expect(Q.unhandledReasons.length).toEqual(0);
+    });
 });