release of USER-CUDA package

HI all - we are (finally) making an official
release of a new USER-CUDA package.
See the 14Aug11 patch.

This is due to considerable work by Christian Trott
(U Tech Ilmenau) - thanks! The initial release
includes support for 28 pair styles, 14 fixes,
and 4 computes. The package can get impressive
speed-ups if your script can run for many timesteps
without invoking a non-GPU-ized fix or compute
or doing output, since in that case all data stays
on the GPU.

The new package is a complement to the original
GPU package of Mike Brown (ORNL). Details
and differences between them are in a new
chapter of the manual: doc/Section_accelerate.html

The short story is that either package can be
faster, depending on your GPU hardware, and
the input script you are running, and the size
problem you are modeling. So you can
now try out both packages. See examples/gpu
and examples/USER/cuda for input scripts
with versions of the same problem for both
packages.

Steve