[cond-optab] update and first call for review

[Paolo Bonzini <bonzini at gnu dot org>]

I meant to start the big assembly comparison run today, but I was busy
and tomorrow I have to leave.  So I'll create the branch before doing
that but after finishing building newlibs.

There were no other target-independent change to make.

I fixed the Blackfin regressions.

I also fixed a few minidetails in a couple other ports (ARM and SHmedia;
too large constraints in cbranch, causing unrecognized instructions later).

Building newlib has the following results:

- fails at the same spot as base for arc and m68hc11;

- passes for stormy16 where base fails (PR39601; will investigate if no
one beats me);

- passes for arm, avr, bfin, cris (v1/v32), crx, e500, fr30, frv,
iq2000, m32c, m32r, mcore, mmix, mn10300, sh, sparc, spu, stormy16, v850

- is finishing now for h8300 and sh64

I still have to post m68k, m68hc11, mips changes but I have them.

I'll create the branch early next week, volunteers for reviewing the
target-independent parts are welcome (for now, only those already posted).

I'll need help bootstrapping on pa, rs6000, s390, alpha, ia64, arm.
I'll try assembly language comparisons there too though.

Paolo