Fixes to asinh(), atanh(), and sinh(): The 'basic' expansions all exhibited...

Fixes to asinh(), atanh(), and sinh(): The 'basic' expansions all exhibited bad cancellation errors near zero (<= 1E-10). This can be easily seen e.g. with x = 1E-30, the results are all zero though they should be extremely close to x. The cutoff values (1E-5, 1E-9) are chosen so that the next term in the Taylor series is negligible (for double). Functions could maybe be optimized to use only first term (x) and a smaller cutoff, just bigger than where the cancellation occurs.