Faster hdf reading

[pmrv]

I'm waiting for pyiron to submit a large number of jobs right now, so I thought I have a look at how to make it faster.

One of the major bottlenecks is calling list_all()/list_groups()/list_nodes() to check what datasets/groups are inside the HDF5 files. In fact roughly 75%(!) of loading a small lammps job is spent in FileHDFio.list_all().

This makes two changes:

1. Eagerly try load read data directly with h5io.read_hdf5 instead of checking first whether it is there and then reading it. This makes a simple read like job['output/generic/energy_pot'] faster by about a factor 5, so that it takes roughly the same amount of time as calling h5io.read_hdf5 directly on a file.
2. Use list_all() instead of list_nodes() and list_groups together. This saves opening the HDF5 file once.

Both together make loading a lammps jobs about 10% faster.

I want to mention that this

h5io.read_hdf5(j.project_hdf5.file_name, title=j.name + '/output/generic/energy_pot')

is still about twice as slow as directly doing

with h5py.File(j.project_hdf5.file_name) as f:
    a = f[j.name + '/output/generic/energy_pot']

so there's still room for improvement.

[jan-janssen]

Looks good to me