docs: Add documentation for writing the HW-accelerator operator.

[xyzsam]

This goes through all the steps:

1. Creating scratchpads for local use.
2. Tiling tensors with the tiling library functions.
3. Writing a loop nest.
4. Modifying code to work with gem5-aladdin APIs.
5. Modifying code for LLVM-Tracer.

This does not include instructions for actually running a simulation. That's more broadly useful information which should go somewhere else (maybe on the Python side).

[xyzsam]

ping