ocl_pow selects incorrect built-in function (Bug #3977)

modules\core\src\mathfuncs.cpp, ocl_pow function.

There are a couple of issues in the following code:
bool issqrt = std::abs(power - 0.5) < DBL_EPSILON, nonnegative = power >= 0;
const char * const op = issqrt ? "OP_SQRT" : is_ipower ? nonnegative ? "OP_POWN" : "OP_ROOTN" : nonnegative ? "OP_POWR" : "OP_POW";

'nonnegative' here refers to power, and:
- For integer power either pown or rootn built-in function is selected. The choice of rootn is simply wrong - this function calculates x^(1/y) which is not what should be computed here.
- For real values of power powr or pow is selected. But per powr function specification - 'Compute x to the power y, where x is >= 0', so powr should be selected depending on the sign of base (x), not the power (y).

So this code should either completely remove 'nonnegative' logic:
bool issqrt = std::abs(power - 0.5) < DBL_EPSILON;
const char * const op = issqrt ? "OP_SQRT" : is_ipower ? "OP_POWN" : "OP_POW";

or some additional logic should be added to check the sign of the input elements (InputArray _src) and select powr for non-negative values.