M300433/number conversion improvement

When implementing the conversion functions std::numeric_limits::quite_Nan() was used. Since this Nan is only defined for floating point types the implementation was wrong. When adding the tests contained in this PR this was found and fixed.

• added tests for handling of empty strings
• split number conversion into two functions (integral/floating)
• added exception on wrong conversion
• the two new functions catch this exceptions and return ::max() for integral and ::quite_Nan for floating point representations