June 2019 Training reaction fixing hackathon

[mjohnson541]

trafficstars

Hackathon had two objectives:

1. Remove unreasonable high rates that prevented us from setting a bounds on rates within each family for filtering
2. Remove unnecessary duplicate reactions from the database

Emergent considerations: Removing GAV estimate training reactions Removing training reactions with small temperature ranges

Once this PR is ready to be merged remember to remove the filter threshold caps in RMG-Py. The cap values are the old filter threshold values and introduced to avoid slowdown of RMG if the "high-pressure" Arrhenius fits are unreasonably high based on the current RMG-database master.