- calculate ticksTotal instead of ticksMean
- local / global width is based on ticksTotal value
- added instrumentation for OpenCL program compilation
- added instrumentation for OpenCL kernel execution

2. fixed parsing of "cat[range_spec]ord[range_spec]" type specification string when using ml::TrainData::loadFromCSV(). Thanks to A. Kaehler for reporting it

  * 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

All of these: (performance) Prefer prefix ++/-- operators for non-primitive types.
[modules/calib3d/src/fundam.cpp:1049] -> [modules/calib3d/src/fundam.cpp:1049]: (style) Same expression on both sides of '&&'.

cosmetic changes to fix this warning.
real bugs not found

- #3771 - inline round on ARM
- #5633 - documentation for MSER
- #5666 - run.py fixes

Conflicts:
	modules/ts/misc/run.py

`has()` only tests for an argument presence which is always true for
arguments with default values. Use `get<bool>()` to check the value
instead.

(cherry picked from commit 47cee871)

Fix for bug #5007: moved definition of Size_MatDepth_t and Size_MatDepth from ts_perf.hpp to perf_channels.cpp. This way they are closer to where they are needed and live in a different namespace (possibly the reason why the fix works).