cv::Mat doesn't have width or height methods, but has cols & rows
instead.

It seems that the act of polling itself can cause the device to heat up
so do it only every 3 seconds instead of every 1 second.

Some compilers appear to have difficulty optimizing the previous code so
it can impact performance.

Cut-off point changed when to call the multithreaded implementation in
the OpenCV HAL.

For benchmarks with a single parameter parentheses are not needed.

Benchmark suite from OpenCV's video module was added to the benchmark
list, but the related perf test binaries were not pushed by push.sh

This brings the performance profile in line with OpenCV's which enables
multithreading for these operations.

* Verbose mode's output changed to mimic a shell output with the -x
  flag.
* When a device filename is created the OS separator is removed from the
  very beginning of the filename as filenames starting with a '-' can be
  painful to work with in interactive shells.

Mark Horvath authored Nov 05, 2024 and Michael Platings committed Nov 06, 2024

shlex generates an escaping around '&&' which is not understood by some
shells.

clang-tidy appears to search for the .clang-tidy configuration file in
the current working directory before searching the parent directories
of the source file. Previously a dummy .clang-tidy file was placed in
the build directory with the intent of avoiding running clang-tidy on
external projects like Google Test. Unfortunately, since the build
directory was the current working directory at build time this meant
that header files were not checked by clang-tidy. This change fixes the
problem by putting the dummy .clang-tidy file in the _deps directory
that CMake uses for external projects.

To ensure that headers in new directories such as kleidicv_thread are
checked by clang-tidy, make the header filter regular expression match
everything.