- added support for caching of various metadata generated/used by

APARAPI, thus avoiding the need for re-generating them on each kernel
call (or, worst, when a new Kernel is instatiated)
- added support for finding method properties (such as method using
atomics 32, OpenCL mapping, etc.) based on full method signature, not
only name
- added support for OpenCL-mapped methods user-defined, i.e., outside of
Kernel class
- added two OpenCL-mapped methods 'hypot' in Kernel (more to come)
- added support for ForkJoinPool-based JTP execution (this avoids
creation of a high number of threads for ranges having local sizes
significantly higher than the number of system CPU cores)
- added support for skipping local barriers altogether in JTP execution,
when a local size (x * y * z) equal to 1 is detected