Task graph/thread pool experimental API design

[bryanwjy]

DO NOT MERGE

• Task graph type system
  • Task dependencies stored in type system
  • Cycle check can be done at compile time
• Thread pool, similar API to span, e.g. compile time vs dynamic size
• Requirement:

• Futex/atomic

• TODO:

• Design exception handling API for task graph
• Decide if we want to ignore false sharing overhead for barriers
• Design task scheduler
• Integrate scheduler with thread pool