add atomic functions

[psychocoderHPC]

Atomic add and exchange are currently added with a macro function.
It is better to implement inlined functions for that.

[mxmlnkn]

I don't think inlining works transparently here, because we need to have acc as a parameter.

[tdd11235813]

although issue itself is a bit outdated, I just want to add that few atomics like atomicXor and atomicOr are still missing.

Few more notes:

• Device functions have to provide the acc object in the signature when using CUDA/CUPLA functions
• you might need an additional ALPAKA_NO_HOST_ACC_WARNING to suppress warnings/errors like calling device function from host code

    ALPAKA_NO_HOST_ACC_WARNING
    template< typename T_Acc >
    ALPAKA_FN_ACC
    void operator()( T_Acc const & acc, ... ) const
    {
        // ...
    }