[PATCH, 2.4&2.5, 32bit&64bit] new fpu code

Whew!  What an effort to create a patch for 4 arches instead of one!

Anyhow, attached are the complete patches for the new FPU context 
management code, for both 2.4 and 2.5, 32bit and 64bit.

. Get rid of lazy fpu switch.  New algorithm is SMP safe.
. Simplify signal/fpu handling
. Simplify FPU register access
. Unify FPU code between 2.4 and 2.5, and between 32bit and 64bit

Ralf, good enough?

Jun