Allow for estimation of contact fraction breakpoint dates

[seananderson]

• would somewhat slow down fitting since more estimated parameters would go into the ODE, but on balance it would require less fiddling and arbitrariness to setting breakpoints
• this would remove the f_segment ID argument interface and instead take lognormal priors on the dates of breakpoints