Optimize long double routines

[ZERICO2005]

Changes:

• Optimized softfloat by precalculating the sign of the arguments
• Added C23 issignaling to detect signaling NaN's. This allows for some softfloat optimizations too. All NaN's are signaling NaN's except for positive/negative quiet NaN.
• Slightly optimized _dcmp knowing that the two arguments are 12 bytes away from each other on the stack.
• Precalculated the sign for the float64_to_integer routines
• Removed the _debug_funcl macro aliases since the long double routine linking bugs have been fixed in ez80-clang nightly