Implement quantile-based integration ranges in OneFactorCopula family

[Nityahapani]

Seven FIXME comments across onefactorcopula.cpp and onefactorstudentcopula.cpp flagged the same problem: hardcoded ±10 integration limits for the Y-table in performCalculations() and for the M/Z grids in cumulativeYintegral(), with a note to derive them from a cutoff quantile instead.

The hardcoded limits are fragile:

• For fat-tailed Student-t distributions with low degrees of freedom, ±10 may not capture enough tail mass, biasing the computed probabilities.
• For all distributions, there is no connection between the grid bounds and the actual tails of the marginals, making the quality of the approximation opaque.

Fix: replace every hardcoded bound with a value computed at cutoff = 1e-6 (or 1e-7 for the Z moment check) using the appropriate inverse CDF:

OneFactorStudentCopula (both M and Z are Student-t):
ybound = InverseCumulativeStudent(nm)(1-c)*scaleM
+ InverseCumulativeStudent(nz)(1-c)*scaleZ

OneFactorGaussianStudentCopula (M Gaussian, Z Student-t):
ybound = InverseCumulativeNormal(1-c)
+ InverseCumulativeStudent(nz)(1-c)*scaleZ

OneFactorStudentGaussianCopula (M Student-t, Z Gaussian):
ybound = InverseCumulativeStudent(nm)(1-c)*scaleM
+ InverseCumulativeNormal(1-c)

checkMoments() Z grid: InverseCumulativeNormal(1-1e-7)
checkMoments() Y grid: use the already-computed y_.front()/y_.back()
which now reflect the quantile-based bounds

Also clarify the FIXME comment in conditionalProbability() explaining why p < 1e-10 is guarded (numerical stability of inverseCumulativeY at degenerate tail).

Adds #include for normaldistribution.hpp in onefactorcopula.cpp and studenttdistribution.hpp in onefactorstudentcopula.cpp.