clean up RateCollection

[zingale]

RateCollection has gotten quite big and is hard to read now. I think that we can remove some things.

• the get_screening_map function really only needs a list of rates and can be extracted. This is done in PR move some screening utils out of RateCollection #666
• there are a lot of rate caching functions that were added for the sensitivity logic. These also only need a list of rates and can likely be removed to their own module. We can store the cache arrays in their own class. This includes: calc_count_matrices(), update_rate_coef_arr(), update_yfrac_arr(), update_prefac_arr(), evaluate_rates_arr(), evaluate_ydots_arr(), evaluate_activity_arr(), clear_arrays() (done in Pull the vectorized rate evaluation functions into a separate NumpyNetwork #684)
• the NSE stuff can be split off into an NSENetwork (done in separate NSE stuff into its own NSENetwork #675)
• we can change RateCollection to only take a Library -- this makes more sense in terms of the workflow we use for creating networks
• we can make _build_collection a decorator
• the precedence logic and _make_distinguishable() should be moved to Library. This is consistent with the new workflow that a RateCollection is created with a Library that has no duplicate links (done in remove the precedence logic from the RateCollection constructor #673)

[zingale]

the logic here is that the main methods in RateCollection should be the ones that I user would call directly.