Hash computation seems not unique (across systems?)

[felixhekhorn]

The purpose of computing hashes inside banana is to boil down the full information to make it traceable - however, it seems that the hash is not unique across systems as was revealed in NNPDF/eko#241. At some point I had even locally the problem that the navigator comparison was empty, but the hash was still different.

For the case at hand we had:

• ocard=cc12964 in the CI
• - obs: 8beae75234d8ba49de7079e5f76d3916dd810708d31bb54f85505faf1c470055 locally for myself
• - obs: 5d279b7c4222fc5f9b2858f7438e69a683a0030ea3a41c1093be921e2ae0f25b locally for @giacomomagni

something which just came to my mind: we might of course also depend on the Python version - however, that would not explain my local hickup ...

Transfered from NNPDF/eko#267