and fix up related cruft

Should help to fix the build on OpenBSD (together with a corresponding
patch to libraries/base).

I'm not sure what happened here, but the ordering of tests was messed
up, with the result that the GHC version check was being omitted.

This makes GHC work even if you are actually running it in 64bit mode,
e.g. on OS X 10.6 Snow.

Instead, we just put the flags we need in there (e.g. -m64 on OS X 64).
This fixes a problem found by Simon M, where we were compiling
everything with -g, leading to a bloated RTS.

The build system should use 'find' and 'sort' that are discovered by
configure, not the ones in your path.  On Windows the ones in your path
might well be the non-Unixy Windows versions.

This patch fixes the ones I tripped over. There may be more.

 * use --nonet, so xmllint and co don't go off trying to download
   stuff from the web

 * use the http:// reference for the stylesheet, so we don't have to
   search the filesystem for it (should speedup ./configure)

Rather than using the autoconf built-in stuff and mangling it, we
now just ask the bootstrapping compiler what platform we are on.

When doing a port of GHC, you need to specify the platform you are
porting to.

The minimum version of GHC required is now 6.8.

Now configure looks for it as an installed program instead.

Now we have variables for whether or not to build the docbook docs as
HTML, as PS, and as PDF. The configure script output now matches what
the build system will do (except it cannot take account of any
mk/build.mk settings, of course).

We use the GHC version number, as the old runghc one doesn't seem very
useful.

Patch from sthibaul. Fixes trac #2883.

We get caught by
    http://savannah.gnu.org/bugs/index.php?1516
    $(eval ...) inside conditionals causes errors
with make 3.80, so warn the user if it looks like they're about to
try to use it.

The original patch said:

* I made some changes to the way in which eta expansion happens in
  CorePrep, mainly to ensure that *arguments* that become let-bound
  are also eta-expanded.  I'm still not too happy with the clarity
  and robustness fo the result.
  
Unfortunately this change apparently broke some invariants that were
relied on elsewhere, and in particular lead to panics when compiling
with profiling on.

Will re-investigate in the new year.