Thermo inconsistencies in training reactions

In H abstraction training, there is at least on example where a reaction is specified in both the forward and reverse directions, but the kinetics are inconsistent with the thermo. This can cause a lot of trouble if using the the thermo to generate reverse rate constants when training the tree since the reverse kinetics of one direction will not match the forward kinetics of the other direction. This is happening for OH + N <=> HN + O, which results in a node with very high uncertainty.

entry(
    index = 625,
    label = "HO-4 + N <=> HN + O-2",
    degeneracy = 1.0,
    kinetics = Arrhenius(A=(6.4e+12,'cm^3/(mol*s)'), n=0.1, Ea=(88.9518,'kJ/mol'), T0=(1,'K'), Tmin=(300,'K'), Tmax=(2000,'K')),
    rank = 2,
    shortDesc = """Added by Beat Buesser""",
    longDesc = 
"""
Added by Beat Buesser, value for reaction: N + OH = NH + O (B&D #26b) in 'Gas-Phase Combustion Chemistry' (ISBN: 978-1-4612-7088-1), chapter 2, 'Combustion Chemistry of Nitrogen', Anthony M. Dean, Joseph W. Bozzelli",

Converted to training reaction from rate rule: OH_rad_H;N_atom_quartet
""",
)
entry(
    index = 629,
    label = "HN-4 + O <=> HO-2 + N-2",
    degeneracy = 1.0,
    kinetics = Arrhenius(A=(1.7e+08,'cm^3/(mol*s)'), n=1.5, Ea=(217.878,'kJ/mol'), T0=(1,'K'), Tmin=(300,'K'), Tmax=(2000,'K')),
    rank = 2,
    shortDesc = """Added by Beat Buesser""",
    longDesc = 
"""
Added by Beat Buesser, value for reaction: NH + O = N + OH (B&D #27e2) in 'Gas-Phase Combustion Chemistry' (ISBN: 978-1-4612-7088-1), chapter 2, 'Combustion Chemistry of Nitrogen', Anthony M. Dean, Joseph W. Bozzelli",

Converted to training reaction from rate rule: NH_triplet_H;O_atom_triplet
""",
)

note: the training indices are different then above because I am using a different branch

tree_cross_val.csv
tree_cross_val.pdf

Start with the two reaction entries shown in the issue, indices 625 and 629, and compare their forward and reverse kinetics with the thermochemistry used for reverse-rate generation. Review the attached tree_cross_val.csv and tree_cross_val.pdf for the uncertainty impact; done means the duplicated directions are thermodynamically and kinetically consistent.