DataSpaceIException exception thrown during initialisation (before main())

I have a problem with our [STIR library](https://github.com/UCL/STIR) that uses HDF5, where on our university cluster, `H5::DataSpaceIException` exception gets thrown before anything happens in our own code. Running in gdb gives
```
$ gdb src/utilities/list_projdata_info
(gdb) break main
Breakpoint 1 at 0x48b7c0
(gdb) r
The program being debugged has been started already.
Start it from the beginning? (y or n) y
Starting program: /SAN/inm/tools/build_cluster/SIRF-SuperBuild/Release/builds/STIR/build/src/utilities/list_projdata_info
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
terminate called after throwing an instance of 'H5::DataSpaceIException'

Program received signal SIGABRT, Aborted.
0x00007ffff60f7387 in raise () from /lib64/libc.so.6
(gdb) info stack
#0 0x00007ffff60f7387 in raise () from /lib64/libc.so.6
#1 0x00007ffff60f8a78 in abort () from /lib64/libc.so.6
#2 0x00007ffff6705a95 in __gnu_cxx::__verbose_terminate_handler() () from /lib64/libstdc++.so.6
#3 0x00007ffff6703a06 in ?? () from /lib64/libstdc++.so.6
#4 0x00007ffff6703a33 in std::terminate() () from /lib64/libstdc++.so.6
#5 0x00007ffff6703c53 in __cxa_throw () from /lib64/libstdc++.so.6
#6 0x0000000000458892 in H5::DataSpace::getConstant() [clone .cold] ()
#7 0x00000000004905af in _GLOBAL__sub_I_H5DataSpace.cpp ()
#8 0x00000000010b1d0d in __libc_csu_init ()
#9 0x00007ffff60e34e5 in __libc_start_main () from /lib64/libc.so.6
#10 0x0000000000493750 in _start ()
```
I found https://stackoverflow.com/questions/51622933/library-conflict-when-linking-in-linux-but-not-osx where there are comments this could be due to the "static initialisation order fiasco".

Reviewing the code, I see
https://github.com/HDFGroup/hdf5/blob/5f9d12601c828c8d99e1c019f1192d68e8ee5471/c%2B%2B/src/H5DataSpace.cpp#L31
https://github.com/HDFGroup/hdf5/blob/5f9d12601c828c8d99e1c019f1192d68e8ee5471/c++/src/H5DataSpace.cpp#L42-L60
https://github.com/HDFGroup/hdf5/blob/5f9d12601c828c8d99e1c019f1192d68e8ee5471/c%2B%2B/src/H5DataSpace.cpp#L75
This seems to rely on the compiler initialising `ALL_ = 0` before initialising `ALL`, but I guess this is not guaranteed.

It seems safer (and simpler) to rely on a static local variable as in
```c++
DataSpace *
DataSpace::getConstant()
{
static DataSpace ALL_ (H5S_ALL);
return (&ALL_);
}
```
I'm not sure if this is entirely safe, as `DataSpace::ALL` is still initialised by the compiler at init time, but as long as `ALL` is only used *after* init, it could be fine. (The alternative seems to be to make `ALL` into a function, but that would break a lot of code).

Of course, this same paradigm for `ALL` is used in many HDF5 classes, so it'd need changing there as well.

PS: I have no idea about the `H5dontAtexit_called` code, so just left it out in my suggestion, but that might be a bad idea.
PS: it seems that `getConstant()` could just as well return a reference, avoiding some extra pointer stuff. Then this is getting close to using a singleton.
PS: I'm linking to `libhdf5_cpp.so`, not the static library.
PS: I'm not a C++ expert, and the literature on global variables is very large. Surely my suggestion could be improved, but it might be better than the current code.