Integration of Initial Guess Generation Into the Fireworks Workflow

[mjohnson541]

Running initial guess generation outside the Fireworks workflow before launching has been a longstanding minor annoyance for small sets of reactions with mostly monodentate adsorbates, but becomes increasingly painful for larger reaction sets with more bidentate species.

This PR integrates the initial guess generation process into the workflow with a firework for each adsorbate that generate the initial guesses and then generates the optimization fireworks and the collect firework that later launches the vibrational calculations.

Additionally, I have parallelized the harmonic optimizations for adsorbate guess generation and TS guess HFSP optimizations with joblib. Even on the test example with only 4 initial guess optimizations, parallelizing on 4 processes gave factors of 2-3x speed up let alone for bidentate adsorbate guess and full scale TS guess HFSP optimizations where we can be doing ~80 optimizations.