Revamping the HDF5_VERSION logic.

[WardF]

Revamping HDF5_VERSION logic flow in CMakeLists.txt. Originally, this information was only used to determine if we had met the minimum supported version threshold. It has grown since then. We now check for versions as a cmake-infrastructure provided environmental variable; if that is not available, we attempt to determine the version programmatically using code contributed (I believe) by @edwardhartnett. If we ultimately cannot determine the HDF5_VERSION through any method, cmake will exit with an error.

[edwardhartnett]

The 1.8.22 version of HDF5 works fine for me in CI testing with netCDF.

[WardF]

@edwardhartnett Do you mean 1.8.22 works for you in regards to the HDF5_UTF8_PATHS toggle in this block? Or do you mean in general? Taking a look back, the HDF5_VERSION_REQUIRED "minimum version gate" is set to 1.8.10. Our CI tests against 1.8.21, I'll take a look at bumping that to 1.8.22. I've just noticed a path by which HDF5_VERSION might not be checked to see if it meets the required version, so I'll be pushing that change as well.

[dopplershift]

@WardF this is working in conda-forge/libnetcdf-feedstock#130 for the main build. Still getting a problem with nc_test4:

CMake Error at nc_test4/CMakeLists.txt:50 (IF):
  if given arguments:

    "VERSION_GREATER" "1.10.0"

  Unknown arguments specified

[WardF]

@dopplershift Thanks. I see no errors on the Windows system I'm building against, I'll go dive into the feedstock recipe and see if there is some useful information there for debugging/replicating this.