Implement re-ordering for PLS atomic draws

Atomic mode needs a barrier between overlapping draws. Implement re-ordering of non-overlapping draws to minimize barriers.

Add a massively SIMD "IntersectionBoard" class that assigns a minimal "groupIndex" to each draw based on its bounding box. Sort draws based on groupIndex, plus lower priority items like draw type and texture hash. Issue draws in the new order with barriers between different groupIndex values.

Discard draws ahead of time that are empty or offscreen. (IntersectionBoard doesn't support these.)

Now that the draw logic is getting more complex, store PLSDraws in a smart pointer that guarantees we never miss a call to releaseRefs().

New simd features:
  - reduce_add, reduce_min, reduce_max, reduce_and, reduce_or
  - >2 arguments to join
  - zip()
  - more typedefs

Diffs=
d67aeac4d Implement re-ordering for PLS atomic draws (#6417)

.rive_head

@@ -1 +1 @@
-26612a9523ba4341bf30e30fb479804a98661972
+d67aeac4d7643cdb60f7d22c9b2c0005e7838ed9