Improve reporting of clusterJ test errors.

ClusterJ MTR test runs a jvm to run junit style tests.
If the junit tests have a failure then the test fails
using the MTR/mysqltest --die command.
There is no .result file for the clusterj (or jtie)
tests, so die or exit are the two possible outcomes.

This results in a very minimal message being output for
the failure, and when debugging in PB2, requires that
vardirs are downloaded to see more useful log messages.

To improve this, clusterj and ndb_jtie are modified
to have a recorded result file.

The result file is minimal and does not contain the
JUNIT output.

However if there is a failure running the JUNIT tests
then the last 200 lines of JUNIT output is included
in the test result.

This causes a result content mismatch (as opposed to
the test dieing) which should give more context from
e.g. the PB2 web interface.

Changes made :
 - Make normal output more stable
   - Don't include command binary names in output
     of exec_print_on_error (windows has .exe suffix)
   - Don't include command arguments in output
     of exec_print_on_error
   - Don't include jar names and paths in
     successfully-found paths
   - Don't include jvm path + version in
     success path output
   All of the above give a stable result in the
   success path
   In error paths, most of this info is made
   visible by logging the full java command
   line.  The info is also available in files
   in the VARDIR.
 - Change skip to --skip
 - Modify run_java.inc to not use --die, but
   rather change its output
 - Modify clusterj.test to not use --exit
 - Add a clusterj.result
 - Modify ndb_jtie.test to not use --exit
 - Add a ndb_jtie.result