GMP 4.4 remaining tasks

[Torbjorn Granlund]

After a period of intensive work by Marco Bodrato, Niels Möller and
myself, the number of remaining items for GMP 4.4 is decreasing rapidly!
See <http://gmplib.org/devel/GMP44.html>.

There are just 6 high-priority items (marked in red) remaining:

 * Rewrite mul.c. Keep sqr call for now, but document mpn_sqr_n and
   mention in mpn_mul that mpn_sqr_n should be used.

 * Write proper invert.c. Should use FFT wraparound, and optionally David
   Harvey's mulmid. Perhaps also the mulmod_bnm1 trick. [40h]

 * Make quotient-only division take average time M(log(Q)) where Q is the
   quotient. The current mpn divappr functions require the divisor to be
   normalised, which isn't clever. Redesign! [40h]

 * Update choice between mod_1 and mod_1p_N.

 * Update choice between mod_1 and modexact_1_odd. 

 * Remove many divrem_1.{asm,S,s} since C code is faster already in GMP
   4.3. (If the top-level x86 or x86_64 divrem_1.asm disappear, fat.c
   needs updating.)

The first three are somewhat difficult, but the latter 3 are simple.

There are also about 20 low-priority items remaining, but several of
these will be pushed to GMP 4.5.

-- 
Torbjörn