Refine AutoDifferentiability

[alecloudenback]

A few related things:

• survival(table,from,to) isn't AD'able because that method only operates on integer arguments. Results in a stackoverflow error.
• At the transition point between partial ages, the deriv goes to zero:

Is there a way to generalize decrement and survival to match the integer-age approach at integral ages but avoid the discreteness in the AD? Does it require making the survival curve parametric? An alternative to Balducci/Constant/Uniform?