Prevent adding /MP flag for MSVC in case any other language than C/CXX will be used (e.g. CUDA)

[SunBlack]

In our project we had currently sometimes issues with equivalent CMake code for the flag /MP. Reason add_compile_options("/MP") will add /MP flag to all compilers - even CUDA. Currently this isn't an issue in PCL, because PCL is using CUDA via find_package(CUDA) and not via modern variant (project(pcl CXX CUDA)), see here.

So currently we have 3 options:

1. Do not change anything, because currently no issue with it in PCL (just upcoming)
2. Use old variant (adjust CMAKE_CXX_FLAGS) and add a comment how it looks like if minimum required CMake version will be increased:

    set(MSVC_MP ${CPUCores} CACHE STRING "Number of simultaneously running compilers (0 = automatic detection by MSVC). See documentation of /MP flag.")
    # Replace CMAKE_C_FLAGS/CMAKE_CXX_FLAGS by add_compile_options($<$<OR:$<COMPILE_LANGUAGE:C>,$<COMPILE_LANGUAGE:CXX>>:/MP${MSVC_MP}>) after miniumum required CMake version is raised to 3.11, see https://gitlab.kitware.com/cmake/cmake/issues/17535
    if(MSVC_MP EQUAL 0)
      # MSVC_MP is 0 in case the information cannot be determined by ProcessorCount => fallback
      set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /MP")
      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP")
    elseif(MSVC_MP GREATER 1)
      set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /MP${MSVC_MP}")
      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP${MSVC_MP}")
    endif()

3. Use old variant before CMake 3.11 and new after (see PR)

I don't like option 1, but I have no preference if we use variant 2 or 3. Variant 2 is shorter, but variant 3 is using more modern CMake code.

[jasjuang]

I like variant 3, but if it isn't a lot of work I think we should try to switch add_compile_options over to target_compile_options because the compile flags will actually be propagated to the downstream and won't be lost in the middle. Also having a per target compile option is better than a global compile option because it is more modular.

[taketwo]

Since variant 3 is already implemented, no reason to not merge it I think. In future target_compile_options should certainly be our goal.