Many minor updates.

Removed trailing whitespace from all .sml and .sig files.

This affects over 900 files and was done using emacs's delete-trailing-whitespace
function in batch mode. Building the system with Poly/ML and Moscow ML seems to
work, so I'm hoping these changes don't break anything. Please complain if
they do!

A general update:
1. the decompiler is now slightly faster, due to new code
abbreviations (terms smaller in internal proofs);
2. the definition of the machine-code Hoare triple
was slightly altered to support instruction caches;
3. the x86 model now includes an instruction cache.

Also, I give up on attempting to support MoscowML for
examples/machine-code/lisp. Both the standard kernel and the
experimental kernel crash while loading files in this
directory using the latest patched version of MoscowML.
Both kernels exit with an uninformative message:

Compiling lisp_equalScript.sml
Linking lisp_equalScript.uo to produce theory-builder executable
Uncaught exception: Chr
Failed to build script file, lisp_equalScript

Did they run out of memory? If so, why does the error message not
say so?

Set svn:ignore properties, and also get rid of use of
TextIO.getOutstream, which Moscow ML doesn't have. I'm pretty
confident that TextIO.output and TextIO.StreamIO.output should be the
same. If I'm wrong, please change this back, and we'll mark the
machine-code example as Poly/ML only for the moment.

A general update. Now lisp_finalTheory produces assembly files
containing the verified LISP interpeteres (in machine code):

arm_eval.s
x86_eval.s
ppc_eval.s

Each of these have been successfully run on real hardware, for
each respective platform.

The proof scripts are likely to only work in the experimental
kernel, due to some unfortuante differences between the two
kernels. Some of these differences are exposed more frequently
now due to recent(ish) changes to the datatype package. Maybe
some of the changes made to the datatype package ought to be
reconsidered?