Optimise the choice of function for the gamma P derivative

The incomplete gamma function is evaluated using the Lanczos approximation. This is evaluated using two alternate expressions when the terms a and z are large:

See Incomplete Gamma Eq. 15 and Eq. 16

The defaults used in the Boost C++ implementation required updating to increase the precision of the function when a=142 for STATISTICS-39. The threshold for large a was lowered from 150 to 100. The switch between the two functions requires optimisation to ensure the threshold is applicable to other parameterisations, e.g. a from 50 to 150.