hdf5_der

[svn-r25719] HDFFV-8930:
When id_wrap fails, it has turned the HDF5 library internal “gut” so mess up
that the miscellaneous (misc) test after the id test, could not proceed at all.
It fails and attempts to do h5close() which attempts to cleanup the library
internals which are beyond “help”.  Eventually it went belly up with abort().
The abort() was the result of assert(…) code turned on in the
disable-production mode.  That might explain why things died in debug mode
only in the windows platform.  (I can’t be 100% sure until I could personal
run the code in the Windows platform to track what the failure is.)

My current fix is to:
1. Tell testhdf5 main() to change “exit(EXIT_FAILURE)” to “_exit(EXIT_FAILURE)”.That go around the problematic code in the at_exit(…) cleanup.

2. Moved the id test to the last test so that when id_wrap() fails, no more
tests are attempted.

tested: jam.