Get some better coverage.

Pipelines are taking too long, so to help, remove the Fedora-based
ppc64le jobs, leaving only Ubuntu to run non-accelerated architectures.

This commit makes it much easier to make the software to become
available beyond just it's source code form.

All related headers and libraries can now be installed where desired and
into a few basic consistent directories. This should allow for easy
packaging in the future.

As well, this includes CMake configuration files that allow for easy use
of the `find_package` call to find and configure use.

Until now, all headers were included indirectly through
target_include_directories calls. However with IDE generators, this has
meant that these files were not included in such solutions, making
things less friendly to developers who use such IDEs.

However, through the use of file sets introduced in CMake 3.23, these
files can now be better included, and as such, new CMakeLists files have
been added to most include directories, and generated export headers
moved to these, and most include directories absorbed by the header file
sets functionality.

In preparation for the ability to export/install the libraries, the
'foe' library is being renamed to prevent confusing name clashing with
generic target names such as 'foe::foe' which can be better used as an
interface target.

Fedora alone is a good enough for new distirbution testing.

Vulkan v1.4.309
OpenXR v1.1.45

Now specializes fully o just address issues, leaving leak detection to
the leak sanitizer run.

Use the latest available ltsc, as the best likelyhood for Windows-based
process isolation being available.

The MinGW build of libsodium does not have AES256GCM acceleration.

Alternative GNU/LLVM based toolchains for Windows.

The previous logic included it in cases which were not MSVC-only
environments, such as in MSYS environments.

In the simulation library, an export macro was hiding in a source file.

The MinGW declaration expects a different pointer type.

After Windows pulls a symbol from a dynamic library, as it is C++, cast
it to the destination type of `void *`.

Now runs Windows builds/tests using a 2022-based image, and both Visual
Studio 2019 and 2022.

The CMake scripts glm is installed with can either set some variables,
or create either an imported target of just headers, or a separate
library.

Change from interacting with ninja directly, to doing so via CMake and
CTest, for better future flexibility.

CMake-Scripts up to 24.08.1
Vulkan-Mini-Libs to Vulkan v1.3.295 and OpenXR v1.0.40 support

The sizes of various items such as keys, overhead and hashes were all
suffixed with 'U' making them all unsigned integers.

Rather, make them have no suffixes so that the define can be more
interpreted by the compiler based on the context.

Update the SHA hash tests to use defines rather than hardcoded numbers.

Support for Vulkan v1.3.285 and OpenXR v1.0.36

New leading-edge tooling means slight changes to the formatter.

Change to the safer variants.

CMake can't find it from just the CMAKE_PREFIX_PATH, and needs a more
specific hint.

Vulkan support to v.1.3.281
OpenXR support to v1.0.33

No longer have access to a machine with Vega 7.

Can now create public public keys from private keys individually,
instead of just already complete key pairs.

The physics subsystem uses GLM experimental features, which now throw
errors during compilation if the above flag is not defined.

The signature being used to verify data should be const.

Vulkan support to v1.3.278
OpenXR support to v1.0.33

Previously, after each task the thread would run a condition-variable's
'wait_for' call for at least 1ms, which could become longer based on the
kernel rate.

Now these task runners will run until they are out of tasks, then only
when no work can be acquired, then fall into a condition_variable's
'wait' call.

On Windows, the default polling rate is 64Hz, which means sleep/wait
calls can take up to 15.4ms before even beginning the waking process.

The `timeBeginPeriod` call means we can set the rate up to 1000Hz,
meaning at most a 1ms time between sleep/wait and wake calls on the
kernel side.