Fix SmileSectionRNDCalculator tail quantiles for smiles with fat wings

[quantales]

Follow-up to #2768.

With fat wings, the fixed ±5 ATM-standard-deviation strike grid can make
invcdf saturate too early. Computing the left-tail CDF as 1 - P(S > K)
also loses precision.

This tabulates log-moneyness against $z = \Phi^{-1}(P(S \leq K))$, with nodes
approximately uniform in z and an adaptive search for the range endpoints.
Each tail uses its own put or call digital, with a relative gap of |K| * 2e-6
to limit roundoff across widely varying strikes.

The nStd default changes from 5 to 8. This also changes
GaussianCopulaSpreadEngine prices, since it uses the calculator's defaults.

Five regression tests cover forward reproduction, quantile saturation,
low-ATM-volatility fat wings, spread put-call parity, and the left-tail CDF
against the analytic lognormal result.

[coveralls]

coverage: 75.477% (+0.02%) from 75.462% — quantales:fix/smilesection-rnd-tail-quantiles into lballabio:master