v5.0-ocl_alpha.2

1. Fixed bugs for running the OpenCL version on a multi MPI node configuration.
2. Reduced number of failed regression tests from 13 to 2.
3. Improved OpenCL kernel performance by ~20%.
4. Added caching support for building OpenCL kernels.
5. For Ewald kernels, the Analytical version is used by default.
6. Updated GMX_ prefix to all OpenCL related environment variables.
7. Fixed OpenCL implementation for Cuda Stream Sync equivalent.
8. Temporary disabled ocl fast gen

First CUDA to OpenCL port

Merge pull request #58 from StreamComputing/asc

Fixed version info in README