Change HDF5 backend of fetch_hitemp

[erwanp]

🎯 Todo

Fetch_hitemp currently uses Pandas's Pytables to generate HDF5s. We should switch to Vaex :

• Faster (see Proof-of-concept in radis-benchmarks)
• Probably less memory intensive during the build phase, which could help on memory-constrained environnement (see 🥇 Challenge : compute H2O on Radis-Lab radis-lab#24)

💡 Implementations

• fetch_hitemp calls hdf2df. Most changes happen there.
• we should maintain compatiblity with Pandas's Pytable for the transition. Also with h5py re-add h5py support in cache_files #239 (still not done). I'd refactor hdf2df and introduce a "backend" parameter. Default Vaex.
• decide if we add a temporary function to convert current .h5 cache files into Vaex's .hdf5 automatically. Could be nice not to have to reload, let's say, the HITEMP-CH4 database.

---

_{Tip : you can vote for New Features on https://feathub.com/radis/radis}