- LogisticRegressionImpl::predict() was changed to return the predicted value and to only write to the OutputArray results if specified (no longer segfaults).
- Refactored batch and mini_batch training to use a common gradient computation function (removed duplicate code).
- Altered the cost computation so that NAN is not computed unnecessarily.
- Greatly simplified (and sped up) the code that appends a column of 1s to the data.
- Minor code cleanup.

Instead of chaining a bunch of sanity checks together with "&&", let's just have several asserts. That way, when an assert fails, you don't get a monsterous "<huge evil expression>
failed" error, but only the bit you care about, making your life rather a lot easier.

Conflicts:
	modules/imgcodecs/include/opencv2/imgcodecs.hpp

set GA_Update flag in order remove ERROR 8: band 1: Attempt to write to read only dataset inGDALRasterBand::Fill()

Current implementation returns NaN for some configs of FFMPEG, call default get_fps() instead (as also used by PROP_POS_MSEC)

build for some OCL drivers.
Some OCL vendors treat implicit scalar-vector conversions for
operators as errors when type conversion is required.

OpenBSD needs m and pthread like Net- and FreeBSD.