[flang] [llvm] [flang][runtime] Replace recursion with iterative work queue (WORK IN PROGRESS) (PR #137727)

[Peter Klausler via llvm-commits]

klausler wrote:

The use of the work queue is now optional in this patch.  It's on by default for the GPU, off by default for the CPU.  This should minimize the overhead of this approach when it's not needed.  (Although good performance measurements remain to be done on real applications, and it's not clear what the consequences of this framework are, enabled or not.)

Reviewers: I'm unlikely to make further major changes here until you have comments.  It's stable and working in both modes on the CPU.

https://github.com/llvm/llvm-project/pull/137727