Improved integral robustness

Some integrals used to erroneously generate NaN or Inf values when the boundary coordinates of two basis functions are too close to each other. In this release, we there lower the "zero" for double precision floating points from 1e-10 to 1e-12 and add additional checks for analytical integrals.

The more stringent zero definition (=1e-12) brings a better capacitance accuracy compared to reference values (have observed more than 1% accuracy improvement). The additional checks have a very slight impact on extraction performance so we decide to set this feature on by default. The flag is ROBUST_INTEGRAL_CHECK in caplet_solver/include/caplet_parameter.h.