Move reduction lowering from DistributeOp to TeamsOp and use teams reduction clauses to generate info.

[jsjodin]

This patch moves the reduction lowering from the DistributeOp to the TeamsOp by enabling the reduction clauses on the TeamsOp and generate the reductions from those clauses. This makes the composite code generation obsolete since the reduction information can be recomputed. This change should also allow having separate teams and wsloop reductions in the same block.
Co-author: @skatrak

[skatrak]

Thank you Jan, this indeed looks a lot better. I have a couple of nits, but generally LGTM. If it's already passing Lit and smoke-fort tests I think we should merge it.

[skatrak]

Nit: Maybe we can name this test "reduction-target-spmd.f90" and keep another simpler "reduction-teams.f90" with the same example that was removed from the "Todo" subdirectory.

[skatrak]

Nit: Remove HasDistribute function argument, since it's no longer necessary.

[jsjodin]

Yes, that's a good point.

[skatrak]

Suggested change

	bool isNowait = false, bool isTeamsReduction = false,
	bool hasDistribute = false) {
	bool isNowait = false, bool isTeamsReduction = false) {

[skatrak]

Nit: We can remove this comment now, since we're no longer looking for the wrappers of an omp.loop_nest.

[skatrak]

Don't we currently need this for other reductions? If not, do we expect to use it elsewhere or is this actually only needed to be calculated for teams reductions within a target region?

[jsjodin]

I think it is only needed for the team reduction, but I figured if we ever want to know the data size for parallel/workshare it doesn't hurt to parameterize the code.

[skatrak]

Nit: I think the previous formatting is the preferred one (no unnecessary braces and no 'else' after 'return').