Make projections with modified f values use the minimum possible unique f parameters

[seananderson]

Currently the approach is to treat every projection year as having a unique f parameter and they can just happen to be the same value.

This is simplest coding wise to also allow for the flexibility of multiple changes in f in the future (e.g. cyclical patterns) but it slows projections down since it adds a lot of gradient calculations.

Instead the future f values could take segment values as is done in the fitting phase to minimize the number of parameters in the ODE.