Hello,
I don’t believe you can use Impala on Hortonworks, but I could be wrong. I have never seen it. And neither Hortonworks, nor Cloudera would support it.
Something you have not mentioned is using Hive on Tez. This allows Hive queries to be executed with a YARN based engine which is not MapReduce. This is for the most part a simple configuration change in Ambari. I cannot say whether it is as fast as Impala – but it is worth trying out.
I believe Impala is a Cloudera product and not an Open Source one, so the Cloudera version is the only one in existence. Note that it includes C/C++ code for speed and is not pure Java. Even if Hortonworks wanted to include Impala as a component in HDP they do not have the rights to.
In any case, goodluck.