Trying to resolve some discrepancies between 3.7.7 and 4.0.0-alpha. Currently, presolve solution give consistent RTMs, but different converged solutions. Fixes are:

1. L_tot needs to not have eq_11 terms applied to it, or we double count
2. the cache from reader.py was getting polluted from the get_L_coupled call, as the stored dictionary was mutable. To fix, the logic inside get_L_coupled was modified to create new variables during the inline multiplication, rather than write over.