math/big: improve assembly cores: alignment and unrolling

[griesemer]

Suggestions from Torbjörn Granlund (personal e-mail):

"Aligning:

Consider aligning loops to at least a 0 mod 16 address.
Same for function start.

Tail code after unrolling:

Setting up a loop which runs 0-3 iterations is quite expensive.
It should be better to follow the pattern:

    test $2, Rcount
    je       skip2
    restore CF
    [do two "iterations"]
    save CF
  skip2:
    test $2, Rcount
    je       skip1
    restore CF
    [do one "iteration"]
  skip1:
    [footer]

"