[v2 FEATURE]: hpopt learning parameters consistent with train

[akshatzalte]

Currently, for learning rates, chemprop hpopt generates the best parameters as init-lr-ratio and final-lr-ratio. However, for training, we require init-lr and final-lr (along with max-lr) and the ratios cannot be used directly (a small multiplication step is needed to transfer the parameters). A small convenient feature can be allowing init-lr-ratio and final-lr-ratio as arguments to the chemprop train.