Improve transformer performance by skipping unnecessary work about modlauncher

[cpw]

Should be fixed with new forge and modlauncher APIs.

[ichttt]

I just watched some parts of your stream and want to clarify some things:

• The numbers I got where from running forge to the main menu with the code block I mentioned in my forge PR.
• There were actually two optimizations in my modlauncher PR: skipping the whole transformer step if only plugins are transforming (from Line 52 to Line 71) which I think you didn't pick up. This is badly readable in the git diff though.
• I think another optimisation with quite some impact would be to ask plugins for their needed ClassWriter flags, so if there is any transformer preset, use that COMPUTE_FRAMES, otherwise use the highest level that resulted from the plugins that actually wrote (as e.g. the object_holder_definalize needs no writing flag). Could be a new method on ILaunchPluginService. Currently this is hardcoded here

Anyways, thanks for doing this the proper way and making actual benchmarks. Hopefully I can join next stream