[important] Fix stacktrace in unread.js.

There is a scenario where we call process_read_message()
for a message that we haven't recorded as unread before.
I'm not sure how it happens, but I put back code to
guard against crashing.  The regression happened in
5752458c821.

(imported from commit 5ce15d2e236b738b445ed88f1733aa0612be0ff3)

static/js/unread.js

@@ -72,7 +72,13 @@ exports.process_read_message = function (message) {
     if (message.type === 'stream') {
         var canon_stream = stream_data.canonicalized_name(message.stream);
         var canon_subject = stream_data.canonicalized_name(message.subject);
-        unread_subjects.get(canon_stream).get(canon_subject).del(message.id);
+        var stream_dict = unread_subjects.get(canon_stream);
+        if (stream_dict) {
+            var subject_dict = stream_dict.get(canon_subject);
+            if (subject_dict) {
+                subject_dict.del(message.id);
+            }
+        }
     }
     unread_mentioned.del(message.id);
 };

zerver/tests/frontend/node/unread.js

@@ -215,6 +215,19 @@ var zero_counts = {
 
 }());
 
+(function test_phantom_messages() {
+    var message = {
+        id: 999,
+        type: 'stream',
+        stream: 'foo',
+        subject: 'phantom'
+    };
+
+    unread.process_read_message(message);
+    var counts = unread.get_counts();
+    assert.equal(counts.home_unread_messages, 0);
+}());
+
 (function test_private_messages() {
     narrow.active = function () {
         return false;