Loop versioning

This patch contains code to perform a loop versioning
optimisation phase. Loop versioning replicates loop
bodies and removes from one of those loop bodies
runtime exceptions that can be explicitly tested. The
phase runs as an SSA optimisation so as only to
consider true dependencies. To capture the loop
structure the phase uses an annotated LST node. The
annotations capture information about the initial and
terminal loop iterator values, as well as the loop
iterator instructions.

The current loop unrolling optimisation phase creates
loops unsuitable for versioning. However, disabling
loop unrolling results in bug 1363830 "Lack of loop
unrolling causes stack overflow". For this reason the
loop versioning optimisation is disabled in this patch
but can be enabled with the switch
-X:opt:loop_versioning=true.

The performance benefit of the patch on SpecJVM
mpegaudio (on my Athlon XP 2600+) is to go from ~4.5s
to ~4s on the best of 10 runs with a development build
and the rvm flags of -X:aos:initial_compiler=opt
-X:irc:O2. There's more data in
ftp://ftp.cs.man.ac.uk/pub/apt/papers/JishengZhao_PREP05_Paper.pdf,
however, I've tweaked the optimisation yet further from
there.

   STATEMENT OF ORIGIN FOR MULTIPLE CONTRIBUTORS

We, Ian Rogers, and Jisheng Zhao:

(a) represent that either:

(i) We are the only authors and owners of the
contributed software
(described as/entitled Loop Versioning),
which was neither derived nor copied from any
other software,
or

(ii) that any exception to (i) is software which was
obtained under the
CPL (Common Public License),

and

(b) hereby agree to license this contributed software
under the CPL.