Add optimization compiler options

[RenatoUtsch]

Add options to enable/disable the OpenCL/RenderScript compiler optimizations.

On RenderScript, there are three optimization levels:
#pragma rs_fp_full - No optimizations, full IEEE 754-2008 compliance
#pragma rs_fp_relaxed - flush to zero denorms and round towards zero optimizations
#pragma rs_fp_imprecise - rs_fp_relaxed, NaN and +-inf operations undefined and -0.0 can return +0.0 instead optimizations.

OpenCL has optimization flags that can simulate the three RenderScript levels, and can be specified in the ParallelME::Program() constructor (the options of lower levels are preserved in each new level):
-cl-strict-aliasing - full IEEE 754-2008 compliance, assumes strict aliasing
-cl-single-precision-constant -cl-denorms-are-zero - flush to zero denorms and treat double precision constants as single precision optimizations
-cl-fast-relaxed-math -cl-no-signed-zeros -cl-mad-enable - NaN and +-inf operations are undefined, -0.0 can return +0.0 and mad optimizations

Some OpenCL optimizations made the compiler crash. This needs to be tested.

[wcmjunior]

For future reference:

This is the optimization that is currently being used as default in ParallelME runtime: "-Werror -cl-strict-aliasing -cl-mad-enable -cl-no-signed-zeros -cl-finite-math-only".
And this is the default in RenderScript: #pragma rs_fp_imprecise