Simplify only required functions and not the whole program

[harendra-kumar]

Currently, after the mark inline phase we run the simplifier on the whole program. Can it have some undesired consequences and slow down the code? Can we run the simplifier only on those functions that we want to fuse instead of running it on the whole program?

[harendra-kumar]

Currently, the fileio benchmark shows some regressions when run with fusion-plugin. The larger goal is to make sure that fusion-plugin does not cause any regressions at all so that we can run it without any caution/warning.