Add "teraread" example

Teragen is a good benchmark of raw DFS write throughput. Terasort is a good benchmark of the whole MR system (input, shuffle, output). I've added a simple "teraread" example which reads through the terasort input data without performing any processing: this acts as a good benchmark of a read-only workload (similar to real-life "find a needle in a haystack" MR jobs)