random number issue

[Nikodemus Siivola]

  RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
  RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
  the mean of the distribution can be systematically O(0.1%) wrong.
  Just increasing R-I-E-B is probably not a good solution, since
  it would decrease efficiency more than is probably necessary. Perhaps
  using some sort of accept/reject method would be better.

[Nikodemus Siivola (nikodemus)]

Seems to still hold.

[Lutz Euler (lutz-euler)]

As promised on sbcl-devel, here is the first version of a patch for review.
I have some questions about it on my own which I will post to sbcl-devel.

[Lutz Euler (lutz-euler)]

I have just committed a reworked version of the patch from #2.
Differences:
- renamed "random-bignum.lisp" to "bignum-random.lisp"
- renamed RANDOM-CHUNK-LENGTH to N-RANDOM-CHUNK-BITS
- rewrote various comments to explain things better;
  use "accept-reject" instead of "accept/reject" throughout
- corrected a factual error in the commit message and a comment:
  %random-fixnum uses only one random chunk when sufficient
- added a NEWS entry
- added credits to the commit message