See comments for details

Including help for directory-specific targets, such as 'make 1' in ghc

i.e. not dph, and hence not stage2

i.e. not dph, and hence not the stage 2 compiler.

Fixes failure when Haddocking Data.Monoid in libraries/base

We now also test runghc and unlit

Now that base does -liconv when it is necessary, it is no longer
necessary for haskeline to do it as well, as haskeline depends on base.
Thus we don't need the haskeline-specific hacks in ghc-cabal any more.

test is tcfail205

We needed some more $s to delay evaluation until the values are
available, and the calls needed to be later in the ghc.mk so that
compiler_stage2_WAYS etc are defined.

They now live in the integer-gmp package.

This is a correction to the patch:
* When linking a shared library with --make, always do the link step
which used the wrong flag in making the decision. It used -dynamic
whereas the correct flag is -shared.

It's already checked for foreign import, but was missing for export.

No longer need them as temp vars in the cmm primop implementations.

Using global temp vars is really ugly and in the threaded case it
needs slots in the StgRegTable. It'd also be pretty silly once we
move the cmm primops out of the rts, into the integer-gmp package.

It using the mp_tmp_w register/global as a convenient temporary
variable. This is naughty because those vars are supposed to be
for gmp. Also, we want to remove the gmp temp vars so we must
now use a local stack slot instead.

This means that, on Linux, we get functions like gamma defined when we
#include math.h

With the exception of GHC's main Parser.y(.pp), which has 2
reduce/reduce conflicts