ts dependency from CUDA runtime removed. All implicit CUDA calls replaced by calls from core module.

if `Scalar:all` is used, Mats with 5+ channels cause errors

Because apparently I love running performance tests for debug builds.

Looks like we don't actually use them; when creating precompiled headers
with Visual C++, a dummy .cpp is created at build time.

That way, core/internal.hpp doesn't have to depend on cvconfig.h, which we
don't ship.

Now it allows choosing between arbitrary implementation variants.

Namely, normalize their names to a common convention and remove useless text
from their values.

It is already marked as a gtest 1.7.0

cmpEps replaced on cmpEps2 to reduce code dublication;
Constants for cmpEps return values added.

Use min/max bounds instead of local value to calculate acceptance threshold.
Threshold based on local values somethimes does not work because cancellation
of big values may produce error bigger than local value.

- add syntetic images generation function to ts
- use generated syntetic images