fix lgamma test on i386 and use symbolic constant; ok martynas@

Adapt tests for extended-precision.

Add some simple test cases for lgamma, and lgammaf.
Currently the functions return correct results for the input of
negative zero; however signgam is wrong in this case.