add MallocMC to PMacc using fetchContent

[ikbuibui]

The current pattern fails if there are multiple passes of configure and mallocMC was not initially installed.
This happens because in the first pass add_subdirectory creates the package config and in later passes findPackage finds this and fails as mallocMC wasnt actually installed.
Switched to simply using FetchContent with FIND_PACKAGE_ARGS which first checks with findPackage and then falls back to the source dir which is exactly what we want.
https://cmake.org/cmake/help/latest/guide/using-dependencies/index.html#fetchcontent-and-find-package-integration

[PrometheusPi]

looks good

[chillenzer]

Pertinent infrastructure from mallocMC:

• Usage in CMakeLists.txt: https://github.com/alpaka-group/mallocMC/blob/db9742e0926df6aa38b09b3589b3d49b6f0c87ef/CMakeLists.txt#L41
• corresponding CMake function:: https://github.com/alpaka-group/mallocMC/blob/db9742e0926df6aa38b09b3589b3d49b6f0c87ef/cmake/add_controlled.cmake
• Package information: https://github.com/alpaka-group/mallocMC/blob/db9742e0926df6aa38b09b3589b3d49b6f0c87ef/cmake/package-lock.cmake
• PR introducing proper CMake support: Update cmake alpaka-group/mallocMC#275

There is an on-going discussion with @psychocoderHPC about pulling up the same approach to PIConGPU (namely add_controlled). I wouldn't let the perfect be the enemy of the good but we should consider and be explicit about how this PR relates to and affects a major overhaul in this regard.

[psychocoderHPC]

Not sure but I think we can not use a pure fetch content because on clusters without internet we need at least the way to use an installed version of mallocMC. IMO this is not possible via fetch content.
@chillenzer used CMP in mallocMC to allow selecting the behaviour how dependencies will be provided. fetch from web, used pre installed, ...

[ikbuibui]

The fetch content here is using 'SOURCE_DIR' without a download method, so it doesnt need internet. It expects mallocMC to be at the pointed directory.
But before this, first it is checking for an installed mallocMC because of the set 'FIND_PACKAGE_ARGS'.

This is exactly the same workflow as before.

[PrometheusPi]

@psychocoderHPC and @ikbuibui What is your final verdict on the matter of download needed or not?

[ikbuibui]

We will further discuss our favoured approaches in the office and then decide on how to proceed.

[ikbuibui]

Discussed offline with @psychocoderHPC :
For now this change is okay since it is only fixing a bug while maintaining the status quo, and not changing functionality.
What an ideal solution would be is what @chillenzer was pointing to from his work on mallocMC (add_controlled), which will give the user the option to be explicit on how they want to deal with dependencies.
It is currently implicitly assumed in the cmake code that the we should try to use an installed version first and if that is not available, we will use the subtree. In the future. this implicit decision should be made explicit and provided to the user in a convenient way.