Reshuffled animsys_evaluate_fcurves so it reuses the last resolved rna path.
Localized small float/int/boolean arrays to reduce the RNA read/write overhead.
Values are written to the rna when switching to the next RNA property.

This patch does not sort the fcurves but most likely they are already sorted.

Impact on performance increase is from 12.4 fps to 12.5 fps using Spring 02_020_A.anim.blend on a Ryzen 1700.