Implement asynchronous allocation of components in derived typed coarrays and caf_is_present

[vehre]

Pull request #259 adds the changes needed to opencoarrays/caf_mpi to "work" with current gcc-trunk again. By "work" I mean compile and run the existing tests.

My idea for testing async alloc of components:

check_remote_alloced_comp_1.f90.txt

fails to run for two reasons:

1. in caf_register() the MPI_allocate() syncs and
2. caf_is_present() is not implemented.

The test case can be used as a guide for implementing the features. At the moment it has some additional prints for debugging, which should be removed in the final version.