Registering host/device data from applications

[sayerhs]

This PR introduces data structures for registering host/device data for MeshBlock and CartGrid/Block

• [x] Info data structs for MeshBlock and CartGrid/Block
• [x] MeshBlock
  • [x] Create interface to register data using the new info struct objects
  • [x] Hook up legacy class instances for testing new API using existing code and to allow gradual transition
  • [x] Update Tioga class to register MeshBlock using new interface
• [x] CartGrid/Block
  • [x] Create interface to register data using the new info struct objects
  • [x] Hook up legacy class instances for testing new API using existing code and to allow gradual transition
  • [x] Update Tioga class to register MeshBlock using new interface

[sayerhs]

@jsitaraman

• The current working of tioga_utils is unaffected by these changes. They still use the old registration functions in Tioga class.
• The info object contains both hptr and dptr that are valid and populated by the solver. TIOGA shouldn't worry about that, beyond checking if the pointer is valid (hptr == nullptr) or (dptr == nullptr).
• m_info_device not strictly necessary. It provides a view of the info struct on GPU. So you can do some_gpu_func<<<blocks, threads>>>(m_info_device) and access the data structure the same way in device functions as with host functions. Without this, you would do

some_gpu_func<<<blocks, threads>>>(m_info.xyz.dptr, m_info.iblank_node.dptr, ...);

[sayerhs]

Another clarification... when the solver registers minfo, both hptr and dptr will be valid. On non-GPU builds it will just point to the same memory location, on GPU builds it will contain the host/device views. We will not register a separate minfo_device.