It makes the parsing simpler if we always have the same number of columns

With 30000 I was getting:
    xargs: value for -s option should be < 28153

After rumination, yet again, on the subject of arity calculation,
I have redone what an ArityType is (it's purely internal to the
CoreArity module), and documented it better.  The result should
fix #3959, and I hope the related #3961, #3983.

There is lots of new documentation: in particular
 * Note [ArityType]  
   describes the new datatype for arity info

 * Note [State hack and bottoming functions] 
   says how bottoming functions are dealt with, particularly
   covering catch# and Trac #3959

I also found I had to be careful not to eta-expand single-method
class constructors; see Note [Newtype classes and eta expansion].
I think this part could be done better, but it works ok.

The problem was that a strict binding was getting floated
out into a letrec. This only happened when profiling was
on.  It exposed a fragility in the floating strategy.  This
patch makes it more robust.  See
      Note [Avoiding unnecessary floating]

This bug just led to stupid code, which would
later be optimised away, but better not to generate
stupid code in the first place.

Simon M and I looked at this, and we think GhcLibProfiled is
(a) not needed (b) confusing.

Ian should review.

Really, if GhcProfiled is on we should also 
check that 'p' is in the GhcLibWays

Reason: DPH uses Template Haskell and TH doesn't work 
in a profiled compiler

For deriving Functor, Foldable, Traversable with empty 
data cons I just generate a null equation
   f _ = error "urk"

There are probably more lurking (eg Enum) but this will do for now.

Simply fixing #4203 is a tiny fix: in case alterantives we should
do dumpUDs *including* the case binder.  

But I realised that we can do better and wasted far too much time
implementing the idea.  It's described in
    Note [Floating dictionaries out of cases]

Silly bug really, we were running the C finalizers while the StablePtr
table was still in a partially-updated state during GC, but finalizers
are allowed to call freeStablePtr() (via hs_free_fun_ptr(), for
example), and chaos ensues.

I encountered a couple of things that broke after Ian's previous
patch: one was my nightly build scripts that use 'make stage=2' at the
top level, and the other is 'make fast' in libraries/base, which uses
'stage=0' to avoid building any compilers.

So my version of this patch is more direct: it just turns off the
appropriate dependencies using a variable set by 'make 1', 'make 2',
etc.

"printable" is ASCII-only, whereas in other locales we can get things like
# 1 "<línea-de-orden>"

If we start with a .cmmcpp file then they don't get flattened in
the CmmCpp phase, as we don't run that phase.

as PIC is always on

The default is a new "some" state, which allows only known-safe flags
that we want on by default. Currently this is only "--info".

These make less sense now we support multiple languges. The
"languageExtensions" function gives the defaults.