Add invoke fast-path support for quickened invokes.

Quickened invokes had their own handler implementation.
Merge it into the generic DoInvoke which has the fast-path.

This speeds up arm64 golem interpreter benchmarks by 6%.

Test: test.py -b -r --interpreter --host --64
Change-Id: Icac9e073f61df67780242877179111ed7bee7154