Regression in compound target generation

[SergioRAgostinho]

The PR #3044 introduced a regression in #2673. By turning cmake macros into functions, it modified the previous existing variable scope and broke the compound target generation mechanism.

The current possible options are:

• I believe looking into the parent scope as in set(target_list ${target_list} ${new_target} PARENT_SCOPE) might be sufficient, but prone to break if we rearrange our scripts further
• Use INTERNAL CACHE variables - This persists over runs, so if the user decides to changes the which targets he wants built half way through configuration, this variable will not reflect the most current state.
• Use GLOBAL PROPERTIES - I suspect this one might face the same persistence issue as the INTERNAL CACHE
• revert back from function to macro - These target creation routines are not defining temporary variables inside, so there's no penalty in doing so. Suffers from the same breakage risk as the first option, with respect to future changes.