[FEA] Automatic Packet <-> Ordered GPU Tensor Functionality

[dleshchev]

Assembling packets to tensors feature

DAQIRI should include packet to tensor conversion as a functionality. Getting to that point is a multi-step process that will include the following steps:

• Document/provide example(s) of how to write tensor assembly CUDA kernels for different use cases with/without HDS
• (TBD) Make writing this step agent-friendly. The users should be able to accomplish the task by providing an agent with an example and their specific packet->tensor format.
• Gather packet->tensor examples/use cases to understand commonalities and see how this can be generalized
• Formalize the solution and integrate into DAQIRI

[awthomp]

Resulting tensor format should be:

• In order
• Support community GPU data standards like dlpack and the __cuda_array_interface__