wrong fp16 adder: -2+2=2

[fanfanfan-hff]

a,b:c000 4000
bigger,smaller,aligned_small:4000 4000 400
adder_input_1,adder_input_2,adder_output,400 400 000
result:4000

should be 0

[chamchamgo]

reproduced , maybe normalization code has bug.

[erihsu]

@fanfanfan-hff should consider the case into directed test. any founds of bug will be appreciated and willing to merge request. I believe no one until now use this code into real chip project.