Memory leak in Fortran wrapper for the profile function calls

[tyronerees]

As pointed out by @mjacobse in #57, there is a memory leak in the Fortran wrapper for the profile function calls:

spral/src/ssids/profile_iface.f90

Lines 82 to 102 in c8c3fcd

	nregions = size(regions, 1)
	allocate(f_regions(nregions), stat=st)
	do i = 1, nregions
	f_regions(i)%nproc = regions(i)%nproc
	ngpus = size(regions(i)%gpus, 1)
	f_regions(i)%ngpu = ngpus
	if (ngpus .gt. 0) then
	allocate(gpus(ngpus), stat=st)
	gpus(:) = regions(i)%gpus
	f_regions(i)%gpus = c_loc(gpus(1))
	nullify(gpus)
	end if
	end do
	c_regions = c_loc(f_regions)
	call c_begin(nregions, c_regions)
	! TODO free data structures
	end subroutine profile_begin

From PR #57:
Seems like it should not be too difficult to fix, a quick guess would be something like this:

do i = 1, nregions
   if (f_regions(i)%ngpu .gt. 0) then
      call c_f_pointer(f_regions(i)%gpus, gpus, [f_regions(i)%ngpu])
      deallocate(gpus)
   end if
end do
deallocate(f_regions)

But I was a little irritated that it was left as a "TODO", maybe I am missing something. And I'm not sure if I could test it properly, haven't used the solver with any GPU yet, let alone multiple.

Originally posted by @mjacobse in #57 (comment)