`decimal128_t` fails addition edge cases that 32 and 64 do not

[mborland]

In the edge cases this fails at 128-bit:

dqadd71310 add 1E34  -0.51                ->  9999999999999999999999999999999999 

But the equivalent at 64-bits runs just fine:

ddadd71310 add 1E16  -0.51                ->  9999999999999999      Inexact Rounded