[Flang][OpenMP] Preliminary support for teams reductions

[skatrak]

This PR enables Flang lowering and some preliminary MLIR to LLVM IR translation support for teams reductions.

Translation to LLVM IR is limited at the moment: it has only been tested to work for a simple host-only program.

! test.f90
program main
  integer :: sum
  sum = 0
  !$omp teams reduction(+:sum) num_teams(10)
  !$omp parallel reduction(+:sum) num_threads(2)
  sum = sum + 1
  !$omp end parallel
  !$omp end teams
  print *, sum
end program main

! flang-new -fopenmp test.f90 && ./a.out
! 20

Need to investigate incorrect loop reductions (and most likely target SPMD reductions are broken too).

! test.f90
program main
  integer :: sum
  sum = 0
  !$omp teams distribute parallel do reduction(+:sum) num_teams(2)
  do i = 1, 10
    sum = sum + 1
  end do
  !$omp end teams distribute parallel do
  print *, sum
end program main

! flang-new -fopenmp test.f90 && ./a.out
! 20 (should be 10)

[skatrak]

Closing as it's already merged as part of #159.