Document best practices and available controls for testing of HDF5

[jhendersonHDF]

HDF5 should document known best practices and available knobs for controlling how and where HDF5 can be tested when building the library.

A relevant comment from @gheber on https://github.com/HDFGroup/hdf5/issues/6114, in the context of test failures occurring due to running HDF5 testing on an NFS mount:

Maybe we should give people a short guide like this:

Key knobs for keeping builds/tests off slow filesystems

• CMAKE_RUNTIME_OUTPUT_DIRECTORY, CMAKE_LIBRARY_OUTPUT_DIRECTORY, CMAKE_ARCHIVE_OUTPUT_DIRECTORY, CMAKE_TEST_OUTPUT_DIRECTORY, CMAKE_Fortran_MODULE_DIRECTORY: defaults point to /bin (and /mod for modules) but can be set at configure time to a fast scratch path, e.g., cmake -DCMAKE_RUNTIME_OUTPUT_DIRECTORY=/scratch/hdf5/bin -DCMAKE_TEST_OUTPUT_DIRECTORY=/scratch/hdf5/tests .. (config/HDFMacros.cmake:41-88). CMAKE_INSTALL_PREFIX is also set here; override it if you want the install tree on a different volume.
• HDF5_PREFIX: prepends a path to every serial test file; great for moving serial test I/O to a fast disk (test/h5test.c:50-54). Example: export HDF5_PREFIX=/scratch/$USER/hdf5_serial.
• HDF5_PARAPREFIX: same idea for MPI/parallel tests; critical to steer MPI-IO away from NFS/home (test/ h5test.c:56-61,385-415; release_docs/README_HPC.md:302-334). Example: export HDF5_PARAPREFIX=/scratch/ $USER/hdf5_par.
• HDF5_DRIVER / HDF5_TEST_DRIVER: selects the VFD for tests (test/h5test.c:42-45,2113). Setting HDF5_DRIVER=core keeps test files in memory (with an optional backing store), which can avoid slow storage timeouts; other values like sec2, split, multi, and family change how files are laid out.
• HDF5_NOCLEANUP: when set, test files are left on disk (useful for debugging, but can fill slow disks) (test/h5test.c:370-383).
• Test runtime controls to reduce exposure to slow disks: HDF5TestExpress (or the doc alias HDF5_TEST_EXPRESS) picks quick vs exhaustive suites (test/h5test.c:910-923; release_docs/README_HPC.md:335- 343), and -DDART_TESTING_TIMEOUT=... raises the default 1200s test timeout if a slow FS is unavoidable (release_docs/README_HPC.md:344-350).

Suggested use on slow filesystems: configure with runtime/archive/test output dirs on a fast scratch volume; export HDF5_PREFIX/HDF5_PARAPREFIX accordingly; optionally use HDF5_DRIVER=core for tests and a higher DART_TESTING_TIMEOUT or lower express level if still hitting timeouts.