Re-examine methodHandle methods uninlined by 8144256

JDK-8144256 solved what appears to be a solaris x86-only compiler issue by un-inlining various methods in methodHandle. Reverting that part of the JDK-8144256 reduces the number of cycles/instructions it takes to run a minimal Hello World by ~1.5M, or 1.5% on a minimal test (although static footprint does grow slightly).

Thus additionally, while JDK-8144256 moved to use pass by const reference to avoid use of the copy constructor, startup profiles indicate there are still a number of places where the copy constructor gets exercised, which can likely profitably be converted to use const references instead.