using the LPM_ADD_SUB mega function

example:

"111" + "100" i.e. -1 + -4

111

100

------

1011, discard carry bit : 011 wrong

sign extend first:

1111

1100

--------

11011, discard carry bit 1011 (-5) correct

if I use unsigned adder for signed or unsigned values (and assume negative values are not negative) then that is ok and adder should work without further work.

if I use signed adder then the issue of carry bit has to be taken into account.

Your question why not use unsigned adder always sounds sensible to me but it could be matter of consistency to use signed adder for signed values. I am not sure if it saves much logic using unsigned adder always.