MoLeR inference server hangs on invalid SMILES

[kmaziarz]

If MoLeRInferenceServer receives an invalid SMILES string (e.g. through molecule_generation encode), it just hangs, likely because the corresponding process failed, and the server ends up waiting for it indefinitely. This is far from ideal.

I see two kinds of behaviour we could aim for instead when encoding receives an invalid molecule:

• The top-level process itself also fails with an exception.
• The encoding carries on, returning None for the invalid molecules.

If possible, it could be nice to support both by having __init__ accept an fail_on_invalid_smiles: bool flag.