* check compiler support
  * check HW support before executing
  * add test doing round trip conversion from / to FP32
  * treat array correctly if size is not multiple of 4
  * add declaration to prevent warning
  * make it possible to enable fp16 on 32bit ARM
  * let the conversion possible on non-supported HW, too.
  * add test using both HW and SW implementation

cosmetic changes to fix this warning.
real bugs not found

- Previous version just checked option support, but not enable it.
- Also OpenCV tests reports invalid AVX and AVX2 support status.

1. someMatrix.data -> someMatrix.prt()
2. someMatrix.data + someMatrix.step * lineIndex -> someMatrix.ptr( lineIndex )
3. (SomeType*) someMatrix.data -> someMatrix.ptr<SomeType>()
4. someMatrix.data -> !someMatrix.empty() ( or !someMatrix.data -> someMatrix.empty() ) in logical expressions

Conflicts:

	modules/core/src/stat.cpp

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

Because apparently I love running performance tests for debug builds.

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

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

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