Avoid buffer overrun on encryption

Make sure we null-terminate encrypted strings before passing them to
UTF8ToString.

This used to work when we allocated the buffer on the stack, because it turns
out that allocate() zeroinits the returned memory. malloc(), of course, does
not.

javascript/olm_outbound_group_session.js

@@ -83,6 +83,14 @@ OutboundGroupSession.prototype['encrypt'] = function(plaintext) {
             plaintext_buffer, plaintext_length,
             message_buffer, message_length
         );
+
+        // UTF8ToString requires a null-terminated argument, so add the
+        // null terminator.
+        Module['setValue'](
+            message_buffer+message_length,
+            0, "i8"
+        );
+
         return Module['UTF8ToString'](message_buffer);
     } finally {
         if (plaintext_buffer !== undefined) {

javascript/olm_post.js

@@ -335,6 +335,14 @@ Session.prototype['encrypt'] = restore_stack(function(
             random, random_length,
             message_buffer, message_length
         );
+
+        // UTF8ToString requires a null-terminated argument, so add the
+        // null terminator.
+        Module['setValue'](
+            message_buffer+message_length,
+            0, "i8"
+        );
+
         return {
             "type": message_type,
             "body": Module['UTF8ToString'](message_buffer),