The main goal here is to reduce fragmentation, which turns out to be
the case of #743.  While I was here I found some opportunities to
improve performance too.  The code is rather more complex, but it also
contains a long comment describing the strategy, so please take a look
at that for the details.

People keep complaining, with some justification, that

	runST $ foo

doesn't work.  So I've finally caved in.  The difficulty with the above
is that we need to decide how to instantiate ($)'s type arguments based 
on the first argument (runST), and then use that info to check the second
argumnent.  There is a left-to-right flow of information.

It's not hard to implement this, and it's clearly useful.  The main 
change is in TcExpr.tcArgs, with some knock-on effects elsewhere.

I was finally provoked into this by Trac #981, which turned out, after some
head-scratching, to be another instance of the same problem.

(There was some bug-fixing too; a type like ((?x::Int) => ...) is a polytype
even though it has no leading for-alls, but the new TcUnify code was not 
treating it right.)

Test for this is tc222

Work around the PowerPC architecture's +-32KB limitation for conditional
branches by conditionally skipping an unconditional branch instead
(unconditional branches have a +-32MB range).

This requires an extra pass over the basic blocks for each CmmTop after
block sequencing, to determine which branches are "far".

Fixes ticket #709, "Fixup too large" error with -fasm on PowerPC

conc021 is an example of a program that broke.
It doesn't handle them particularly well still, but it doesn't crash at
least.

Also move closeMutex() etc. into freeTaskManager, this is a free-ish thing

 - Added HPCRIX support for passing tracer filename.
 - Added thread tracing support.
 - Cleaned up use of HsFFI.h