CDH 5.4 had Spark 1.3.0 plus patches, which per the blog post seems like it would not work either (it quotes "strong dependency", which I take means ONLY 1.4.1?). CDH 5.5.x onwards carries Spark 1.5.x with patches. There has been no CDH5 release with Spark 1.4.x in it.
You could use a Apache Spark 1.4.1 release from upstream, manually rebuilt against your CDH5 version of Apache Hadoop, and use the tar-ball paths for all Spark operations, and this should work.
However, such a Spark deployment would not be officially supported by Cloudera Support (if you have a subscription).