Wrap more of the HDF5 API with generated bindings

[jmert]

This wraps more direct uses of ccalls with bindings made with the generator.

There are a few notable exceptions due to either (a) memory handling which the binding generator cannot (yet?) deal with, and (b) a case where a call to the underlying H5Dwrite is relying on being able to declare the buffer to type Cstring for its data validation features.

One more interesting change is to the h5t_get_member_name and h5t_get_tag functions — this changes to using h5_free_memory from Libc.free as noted in the h5t_get_member_name documentation (which says it's more important on Windows).

[musm]

LGTM and a good clean up, just some minor comments!

[jmert]

I added a second commit that moves the definition of h5_get_libversion() and the corresponding use to set the libversion constant. That definition was being used to implicitly initialize the library, so to keep the read_const function working requires an explicit ccall to initialize the library.

It's a bit ugly, but that should maybe just be taken as motivation to work on sorting out the initialization process.

[jmert]

Hmmm... MacOS nightly failed in both Travis and Github Actions, apparently with libhdf5 not being defined, which should be imported from HDF5_jll and works as expected everywhere else (and this PR didn't change that).

[musm]

Strange, let's drop that commit for now so we can get the rest of the changes in?

[jmert]

Strange, let's drop that commit for now so we can get the rest of the changes in?

I've backed off this branch by one commit. Let's see if MacOS CI is still unhappy or not.

Edit: Still fails. Looking at other commits to master, though, and I'm finding that MacOS nightly was already failing there, too — https://github.com/JuliaIO/HDF5.jl/runs/1137069384 — so it's not an interaction with this PR specifically.

[jmert]

Something needs to be fixed in HDF5_jll — I made a branch that just prints a bunch from deps/deps.jl, and it's showing that there is no library on MacOS with Julia latest: https://github.com/jmert/HDF5.jl/runs/1138352787?check_suite_focus=true

[musm]

Thanks for detecting this bug. Let's see if we can get that fixed. So I guess this PR discovered a bug in the HDF_jll wrapper which is good.

[musm]

osx on master should now be fixed

[musm]

want to try that second commit now?