Adds cufft float16 support + option for precision-related error-bound.

[tdd11235813]

• requires half-code library for host-side float16 support (http://half.sourceforge.net, downloaded by cmake)
• cufft results also show available host memory like fftw (clfft client already has it by using OpenCL)
• by float16 overflows are possible yielding to NaNs or Inf in the result, which is treated as an error
  • at the moment overflows happen at 1 Mio elements due to the input data
• results now also contain the number of mismatches, the sample standard deviation of the error
  • appends two new columns to the csv

(results will follow in the gearshifft_results upstream)

[psteinb]

just wonderful! you are putting pressure on me to update my PRs.

[tdd11235813]

I know you are very busy now with the other deadlines. And please don't feel pressured ;) Still not sure, when to go for a new release, but probably after I integrated travis.