fix_radicals() not being called properly

[joegilkes]

When comparing KPM output reactions to Kinetica input reactions, there are some discrepancies with radical reactions. For example:

["[CH]=C", "[CH]=C"] -> ["[CH2]", "[CH2]C=[CH]"] (Kinetica input) | SubString{String}["[CH]=C", "[CH]=C"] -> SubString{String}["[CH2]", "[CH][CH][CH2]"] (KPM output)
["[CH2]", "[CH2]C=[CH]"] -> ["[CH]=C", "[CH]=C"] (Kinetica input) | SubString{String}["[CH2]", "[CH][CH][CH2]"] -> SubString{String}["[CH]=C", "[CH]=C"] (KPM output)

This warrants some investigation, but could be sidestepped entirely by not passing XYZ geometries into KPM and instead using the Kinetica-prepared SMILES

[joegilkes]

Fixed by b35a298. Still unsure why this was happening, but directly using the SMILES from RxData instead of recomputing from geometries ensures consistency, while also massively speeding up predictions by avoiding redundant work and file IO.