#11 Remove Java RuntimeError from stderr of failures

[Tim Yates]

Currently, when Streaming subprocesses fail in Hadoop tasks, the standard error streams contain a stack trace of a Java error complaining that the subprocess failed. This is unnecessary cruft in most situations, since the stderr log will usually contain a message from the subprocess telling why it failed. However, in some cases, like segfaults in C++ tasks, there might *not* be informative error output from the subprocess. An effective resolution would be to replace the stack trace with a concise and clearly demarcated message that the subprocess failed with a particular error code.