Resolve huge numeric discrepancies on different OS and architectures

1. Examples

scatter2d/FindPeaks
scatter2d/Interference1DLattice
scatter2d/LargeParticlesFormFactor
scatter2d/RectangularGrating

do Monte Carlo integration, so relatively big discrepancy is expected.

I increased the number of detector bins, which is important for MC, and the number of integration points.

Exact value of discrepancy is random, changing run-to-run, but its dispersion is common for all OS and architechtures.

2. Tolerances in Suite are back to the default values. No problems are observed.

Resolves #972 (closed)